International Journal of Pavement Research and Technology

https://doi.org/10.1007/s42947-023-00377-w

ORIGINAL RESEARCH PAPER

Investigation of Compaction on Compressive Strength and Porosity

of Pervious Concrete
M. Sajeevan1 · D. N. Subramaniam1 · R. Rinduja1 · J. Pratheeba1

Received: 12 April 2023 / Revised: 2 July 2023 / Accepted: 24 August 2023

© The Author(s), under exclusive licence to Chinese Society of Pavement Engineering 2023

Abstract
Pervious concrete (PC) is a sustainable substitute for conventional concrete application yet limited due to lack of under-
standing on its performance characteristics. The mix design affects the performance, mainly due to its porous structure that
is also not uniform in characteristics when mass produced. Although zero compaction is envisaged for casting of PC, it is
important for mass production of PC with uniform properties. This study analyses the impact of compaction on two primary
performance indicators of pervious concrete, porosity and compressive strength. Laboratory specimens of size 150 mm cubic
were cast with varying aggregate-to-cement ratio (2.5–7.0), compaction (15–75 blows by standard proctor hammer) and
compaction distributions (two types), where water-to-cement ratio was maintained at 0.3 and aggregates used were between
12 and 25 mm. Twelve specimens of each design were cast, and six specimens were tested for compressive strength and
porosity and another six specimens were cored to obtain cylindrical cores of 100 mm diameter for porosity measurements
and porosity distribution analysis using image analytical tools. Results revealed that actual porosity (measured through
image analysis) represented the performance of pervious concrete, and that it is perfectly linearly correlated with effective
porosity. The type of compaction distribution had significant impact on the relationship between porosity and compressive
strength, while the impact was not statistically evident in porosity and compressive strength separately. The performance of
the samples, however, showed correlation to the type of compaction employed when machine learning tools are employed.

Keywords Pervious concrete · Effective porosity · Compaction energy distribution · Compressive strength · Actual porosity

1 Introduction gets affected. Furthermore, there is no heat and moisture

The Industrial revolution contrived multiple detrimental
effects on the environment due to anthropogenic activities
intended to upgrade quality of life [1]. Although in past clay
was used in constructions, following the inclusion of binders
like gypsum and lime in constructions, cement has become
a prominent construction material. This created a revolu-
tion in the construction. Thenceforth, most of the urban
areas are paved with impermeable surface, predominantly
from concrete applications [2, 3]. Due to reasons includ-
ing lack of permeability of conventional concrete pave-
ment, percolation of rainwater and recharge of groundwa-
ter resources are restricted. Since there is an absence of a
persistent water supply to the soil, the microenvironment
* D. N. Subramaniam
daniel@eng.jfn.ac.lk
1
Department of Civil Engineering, University of Jaffna,
Ariviyal Nagar, Killinochchi 44000, Sri Lanka
exchange between the soil and air thereby temperature and
humidity of the urban microenvironment increase [4]. Sub-
sequently, there is an urge to understand significance of the
environment and concept of sustainable development [5–7].
In the construction field the concept of Green Buildings
emerged, that employs several strategies, such as energy
efficiency, water conservation, responsible use of materi-
als and resources, and indoor air quality, to diminish the
negative sustainability impacts of environment [8]. In green
structures, usage of green materials is the most important
criterion. Green materials will have less impact on the envi-
ronment compared to that of conventional materials [9–11].
Pervious concrete is considered as green material for vari-
ous reasons including, usage of more environmental friendly
cements with less carbon footprints, elimination of fine
aggregates, usage of waste material such as fly ash and bot-
tom ash as fillers, reduced material utilization, enhanced per-
meability that promotes percolation of rainwater to recharge
aquifers in urban spaces, reduced heat transfer makes it a

13
Vol.:(0123456789)

best choice for walls to cope with heat and energy losses
related to indoor thermal comfort in buildings [3, 4, 12–16].
Density of pervious concrete ranges from 1500 to 2400
kg/m3, essentially governed by aggregate to cement ratio and
compaction of fresh concrete [13, 17]. Designed aggregate
to cement ratio and elimination of fine aggregates, improves
the porosity of the concrete [18]. Due to higher porosity of
pervious concrete it has lesser strength [13]. The porosity
of pervious concrete varies with design parameters such as
aggregate to cement ratio, compaction, water to cement ratio
and so on [18]. Shape and size distribution of aggregates
also affect the shape and size distribution of pores in pervi-
ous concrete [6, 19]. In addition, it affects the connectivity
of pores and hence the permeability of the pervious con-
crete. Most of the studies analysed porosity obtained from
water replacement method which is commonly referred to as
Apparent Porosity [6, 20]. However, those values depicted
only a portion of the total porosity in pervious concrete that
is inclusive of all accessible pores to the surface and exclud-
ing pores that are disconnected from the surface. Mathemati-
cal models were developed to predict concrete performance
based on the design parameters [21, 22]. Since the pore char-
acteristics are not often considered in mathematical models,
Image analysis is considered as a potential tool that is being
used primarily in the analysis of pore characteristics in con-
crete [21, 23].
Machine Learning is one of the most influential and
powerful tool of Artificial Intelligence which is primarily
used for finding underlying patterns within complex data
and discover new knowledge [24, 25]. There are two major
concepts in machine learning, namely, supervised and unsu-
pervised learning. In supervised learning, we have a set of
labeled data in which we know the structure and outcome
of it, we take this data and train the machine learning model
so it can understand patterns in the data. Once a model has
been trained, we can use it to predict results for data in which
results are unknown. Conversely, if we are given a set of data
that is unlabeled, then we can apply unsupervised to find
patterns that exist within that data [25].
Contemporary literature on pervious concrete research
is centered around laboratory scale experiments and data
analysis using correlation techniques to recognize correlat-
ing concrete design and performance parameters [26, 27].
Since the pore characteristics are not often considered in
mathematical models experimental studies has to be done
before each application, Image analysis was considered as
a potential tool that is being used primarily in the analysis
of pore characteristics in concrete [21, 23]. According to
certain studies that analyzed some pore characteristics such
as porosity spatial distribution and pore size distribution are
with very limited analyses relating to design parameters and
performance parameters [28]. Nevertheless, the ability of
this tool has not been advanced to be employed to improve
13
M. Sajeevan et al.
current understanding on pervious concrete. Moreover, it
was also evident from the contradicting observations from
several studies that there is a perplexity when considering
the impact of aggregate size and aggregate gradation used
in pervious concrete. A particular study used three sizes of
aggregates (2.36–4.75 mm, 4.75– 9.5 mm and 9.5 – 12.5
mm) and concluded that the compressive strength was the
highest and lowest for the largest and smallest aggregates
respectively [28]. Whereas an observation of a certain study
states that there no significant correlation between total
porosity and total strength [29, 30] and another different
study states that there is an increase in compressive strength
in samples with smaller aggregates (2.5–10 mm and 5–14
mm) compared to those of larger aggregates (10–20 mm).
However, the trend was inconsistent in compressive strength
of samples of 2.5–10 mm and 5–14 mm[31].
Total porosity of pervious concrete is a parameter often
discussed in contemporary literature. Aggregate to cement
ratio (A/C), water to cement ratio (W/C), aggregate size and
shape distribution and compaction energy supplied are few
prominent design parameters that affects the total porosity
of pervious concrete. [28, 31–33], Agar [26, 34, 35]. While
increasing the A/C ratio there was a increment in total poros-
ity of resulting pervious concrete while W/C ratio is kept
constant in order to maintain slump of zero (approximately
0.3) [28, 35–37] Due to the decrease in paste volume as A/C
ratio increases (mass of aggregate per unit volume increases
while mass of cement is compromised) there exists a cor-
relation between the total porosity and A/C ratio. The objec-
tive of this study is to analyse the impact of compaction on
compressive strength and porosity and to analyse the pre-
dictability based on conventional mathematical models and
machine learning approaches.
2 Materials and Methods
2.1 Material Selection
Pervious concrete mix comprises of Ordinary Portland
cement (OPC), crushed stones as coarse aggregates and
water. The raw materials were obtained from local dis-
tributor in Northern Province, Sri Lanka and the required
amount of water was obtained through water supply from the
premises, with insignificant chlorine content. The specific
gravity, fineness, and specific surface of the OPC used were
3.15, 10% and ­320m2/kg, respectively. The nominal value
of aggregate size varied between 12 and 25 mm and had
a specific gravity of 2.83. Entire quantity of coarse aggre-
gates required for the study were obtained in one purchase
and retained at room temperature together with a controlled
humidity.
Investigation of Compaction on Compressive Strength and Porosity of Pervious Concrete
The Particle size distribution (PSD) was determined using
image analysis. For this image analysis, chosen aggregates
were drenched in black color oil paint and were laid on white
sheet. An image was taken using a Sony A-7-III camera with
a 50mm /f 1.8 prime lens from a stipulated distance of 350
mm. The particles were then rearranged in another order
and pictured; the same procedure was repeated ten times.
Using the software, ImageJ ™, which is an open software for
processing and analysing images, the obtained images were
transformed to binary scale of black and white (255 and 0,
respectively). The area of each black shape were analysed
that corresponded to aggregates. Thereafter, the diameter
of an equivalent circle of each black zone (area depicting
the aggregate) was computed and categorized into bins of 1
mm, from 5 mm to 30 mm. The cumulative area across 5–30
mm was plotted to represent the distribution of aggregates
size. Average of PSD of aggregates drawn for ten distinc-
tive images taken for various random arrangements of the
aggregates was obtained, as shown in Fig. 1.
2.2 Sample Preparation
A consistent protocol on concrete mixing and total stand-
ard duration that lasted for 20 min was adopted in this
current study. To obtain a measurable slump of zero (0–15
mm) an ideal water-to-cement ratio of 0.28 was used in all
mixes of this study. First, the drum mixer was thoroughly
cleaned by water and dried to eliminate the influence of
remaining water content. Before each mixing, the moisture
content of aggregates were measured to make sure that
the initial moisture content was kept constant throughout
all experiments. The value of measured moisture content
ranged within 1.4–1.8%. Aggregates and cement were
mixed for 3 min, and half of the quantified water was
sprinkled onto the mixture on top of mixing. Subsequently
after another 3 min of mixing the remaining water was
Fig. 1  Particle size distribution
curve of aggregate used in this
study
sprinkled in the similar manner and mixed for another 6
min. The mixture was left at rest for another 3 min during
which the chunks and lumps of material on interior walls
of drum were removed and dropped back into the mixer
drum. In the end the mixture was again mixed for another
5 min and then transferred to concrete moulds in three
layers. A 2.5 kg standard proctor rammer was used for the
purpose of compaction according to the mix design. A
stipulated number of blows (a free fall of 300 mm) given to
each layer. There were two types of distribution of blows
were given to three layers; the first distribution had equal
number of blows for all three layers (x, x and x), while
the second distribution had blows distributed as 1:2:2 (x
for bottom layer and 2 × for other two layers), as shown
in Table 1.
In this experiment a sum of 960 samples were cast by
changing the total number of blows (compaction energy).
15, 30, 45, 60 and 75 blows were given to both compaction
types and several mix designs were obtained. The aggre-
gate-to-cement ratio of 2.5, 2 3.0, 3.5, 4.0, 4.5, 5.0, 6.0 and
7.0 were used throughout this study. All the cubes were
cast adopting a standard cube dimension of 150 mm × 150
mm × 150 mm. Details of all cubes casted is depicted in
Table 2.
Table 1  Types of compaction distributed ratio for 3 equal layers of
concrete cube
Compaction type Ratio of the blows in
each of three equal
layers
Type 01 1:1:1
Type 02 1:2:2
13
 M. Sajeevan et al.

Table 2  Details of cubes casted A/C Compaction (number of was obtained using Sony A-7-III camera with a 50mm /f
with difference mix designs in Ratio blows) 1.8 prime lens from a stipulated distance of 300 mm, under
Experiment 01
15 30 45 60 75
normal lighting conditions and an ISO of 12,800. The cam-
era operating parameters were determined based on trial
2.5 √ √ √ √ √ and error and that yielded the highest quality distinct pixel
3.0 √ √ √ √ √ value. The curved surface was then sprayed with black aero-
3.5 √ √ √ √ √ sol spray paint and dried, which covered all pores, paste
4.0 √ √ √ √ √ and the aggregate surface of the core. Cement–aggregate cut
4.5 √ √ √ √ √ surface of the core samples were carefully painted in white
5.0 √ √ √ √ √ oil paint with a brush. Thus, it was easier to distinguish the
6.0 √ √ √ √ √ pore area from the cement–aggregate cut surface. The core
7.0 √ √ √ √ √ samples were placed on a rotating tray rotating at a con-
stant speed which was controlled by a stepper motor and a
panoramic image was obtained. This photograph covered the
2.3 Testing Procedures entire curvature surface of the pervious concrete cylinders as
shown in Fig. 2.
The weight of every fresh concrete mix in each cube was
measured employing a digital balance with a least count 2.4 Quality Control
of 1 g. Cubes had been then left to dry for a period of 24
h at room temperature after which they were moved to be The entire batch of aggregate was obtained from a single
cured in water for 28 days. The compressive strength of the purchase and stored ambient temperature and at controlled
cubes was measured using the universal compression testing humidity. Required cement was also sourced in one purchase
machine of 2000 kN capacity. Furthermore, cylindrical cores from same brand and stored in a moist free place covered
of diameter 100 mm and height 150 mm were cut out from with polythene wrappers. The quality of the aggregates was
six cubes of each mix. Arrangements have been then made tested using standard testing methods. The moisture con-
to calculate the measured porosity with the aid of using the tent of aggregates was measured prior to each experiment
water replacement method. For this purpose, a glass tank of to check that the initial moisture was the same in all experi-
dimensions 150 mm × 150 mm × 250 mm (length x width ments. Best Laboratory practices were followed throughout
x height) was set up and partly filled with water. The dried the experiment. Prior to preparation of each mix design, the
core samples were immersed in to the tank and water differ- mixing drum was cleaned with water and dried to remove
ence in water level was recorded. Porosity (called measured the dust and other particles which remained during the previ-
porosity) was computed from the difference in water level, ous mixing. Over each mixing day another, three additional
that corresponded to the volume of water displaced. In addi- control cubes were casted to monitor the impact of other
tion, to calculate the apparent porosity, the volume of pores factors on concrete. Similar mixing methods were followed
in samples was calculated based on specific gravity and mass for the control cubes. The weight of the fresh concrete, dry
of solids. Hence, the apparent porosity was calculated by concrete and compressive strength were computed for the
the fraction of difference in entire volume and solid volume control cubes.
divided by the entire volume. In addition, the actual porosity To eliminate errors related to color threshold cube was
of the pervious concrete mixes was also computed utilizing carefully sprayed with black spray paint which covered both
the core cut samples after ensuring that they are completely concrete–aggregate cut surface and pore surface, and dried.
dry, using image analysis method as described below. Subsequently, the cement–aggregate cut surface of the cores
The samples were placed on revolving stage and rotated at was painted with painted white with a brush. This proce-
a constant speed and panoramic image of the curved surface dure made it easier to visibly distinguish the pore surface

Fig. 2  Cropped image and pro-

cessed binary image of A/C 3.0
and 30 blows mix design cube 1

13
Investigation of Compaction on Compressive Strength and Porosity of Pervious Concrete
from cement aggregate cut surface. When the image was
processed in ImageJ™ to convert as a binary image, and
there was no impact from any other colors.
Rather than natural light, annular light was used to main-
tain the same light state for all cubes. Aperture was above
f9 to decrease the blurring effect in pores when the camera
is centered on the cut surface. ISO below 2000 was used to
reduce the noise. All other camera settings were not modi-
fied throughout the study to maintain the same image char-
acteristics. Camera should be fixed in 350-mm distance from
the center of rotating stage and the elevation should be 75
mm from the rotating stage. These precautions will mini-
mize the image error and enhance the image quality.
3 Data Analysis
The images obtained were in color and they were first con-
verted to grayscale (pixel values ranging from 0 to 255) indi-
cating colour from white to black, respectively. This image
consisted of 8 bits per pixel data. Grey color consists red,
green, and blue components with equal intensity in RGB
space, thus a single value indicated relating to each pixel
[38]. Images were then transposed to binary format (black
and white), to easily identify the portion of interests digi-
tally and visually and examined using Matlab, a software
that includes image analysis tools. From the binary images
obtained, the black pixel represented pore surface, whereas
white pixel represented aggregate, or binder cut surface.
Binary images uploaded to Matlab are stacked in a two-
dimensional matrix format containing numbers 0 and 255.
Total porosity in Matlab is computed by the ratio of total
number of black pixels to the total number of pixels.
The images were then split into ten equal layers horizon-
tally and porosity was obtained for each slice. By eliminat-
ing the boundary layers (top and bottom slice) the average
of the other eight slices were obtained and termed as Effec-
tive Porosity (EP). This was done by algorithms written in
Matlab. The data obtained was in a matrix format and was
transferred to MS-Excel and variations were plotted with
depth. Obtained data were also transferred to SPSS, a statis-
tical software for further statistical inferential analyses and
pattern recognition in data variables.
Machine learning classification tools were employed in
this study to analyse the differences between different types
of compaction and EP and actual porosity (AP). Nine dif-
ferent machine learning algorithms were used for the analy-
sis of data that includes, Naïve Bayes, Logistic Regression,
Linear Regression, Support Vector Machine (linear, Radial
Based Functions (RBF) and Polynomial kernals), Decision
Tree, K-Nearest Neighbours, and Perceptron.
Naive Bayes, a classification technique using Bayes’
Theorem in the decision making under the assumption of
independence within predictor variables. This specific algo-
rithm is easy to build as well as useful for analysing large
data sets. Naive Bayes is well known among highly sophis-
ticated classification methods [39, 40]. Logistic Regression
is the other classification algorithm, which estimates dis-
crete values based on predefined independent variables. The
nature of dependent variables in this algorithm is dichoto-
mous which is a binary data coded as 1 and 0 corresponding
to success and failure respectively [41]. The third algorithm
used in this study is Linear Regression which assumes a
linear line to distinguish classes among cases. Here, based
on continuous variables, real values are estimated. The line
obtained from this algorithm that best fits independent and
dependent variables is known as regression line. When one
independent variable is present, Simple Linear Regression
is followed whereas for multiple independent variables Mul-
tiple Linear Regression is followed which could be either
polynomial or curvilinear regression [42].
Support Vector Machine (SVM) is the fourth classifica-
tion algorithm used in this study which is an ML classifica-
tion model that aims to find an optimal hyperplane separat-
ing two different classes (Ben [42, 43]. This study employed
three types of SVMs in this study. Linear Kernel Support
Vector Machine was used for analysis when the data is lin-
early separable. This is particularly useful when there are
large number of features in a particular data set [44]. Follow-
ing it, RBF kernel Support Vector Machine was used, where
RBF kernels are the most generalized form of kernelization.
RBF kernel function for given two points computes how
close they are to each other. The distance between points
is specified in this algorithm which indicates the distance
between the points increases, they are less similar, and the
maximum value could be one indicating that there is no dis-
tance between them [45]. Polynomial kernel Support Vec-
tor Machine is a function that represents the similarity of
vectors in a feature space over polynomials of the original
variables [46].
Thereupon Decision Trees were used, where decision
analysis can be done visually and explicitly representing
choices and decisions. Each branch node represents a choice,
and each leaf node represents a decision [42]. Decision Trees
are simple to understand, interpret and visualize patterns.
Non-linear relationships within parameters could also be
analysed as it does not affect the tree performance. Ran-
dom Forest, which is the next supervised algorithm, is used
in both classification and regression problems in Machine
Learning. Random Forest is a classifier that contains sev-
eral decision trees of given dataset and takes an average to
improve the predictive accuracy of the dataset [47]. Random
Forest Algorithm predicts outputs with high accuracy even
for large datasets.
The other algorithm used was K-Nearest Neighbor,
which is a supervised machine algorithm used to solve both
13
 M. Sajeevan et al.

classification and regression problems. On calculation of
distance between points on graph, KNN learns the similar-
ity among them. This is a simple yet versatile algorithm and
does not require additional assumptions to implement [39].
Perceptron was the next algorithm used which is a super-
vised learning of binary classifiers that decides whether an
input belongs to a specific class or not. Perceptron algorithm
is also best suited for problems that deals with complex data
[48].
4 Results and Discussion
4.1 Porosity and Compressive Strength
Relationships Across Type of Compaction
Distribution
The measured porosity (MP) of pervious concrete sam-
ple was initially calculated. This computation was done
by employing a water replacement method as discussed in
methodology. This porosity value, however, accounts only
for the volume of pores that are accessible to any surface of
the cube and excludes all pores that are disconnected from
the surface. Thereafter, the AP of the core cut pervious
Table 3  Values obtained of arbitrary constants obtained from curve
fitting for AP and MP with strength for both the types separately
Type Plot σ0 B decreased with A/C ratio for all mix designs. In addition, AP
1 AP vs strength 50.17 0.04973
1 MP vs strength 40.13 0.05956
2 AP vs strength 29.9 0.03615
2 MP vs strength 23.59 0.03551
concrete cube samples were computed. This was done by
image analysis as explained in methodology.
For all the cores cut samples the MP and AP values were
computed and compared with compressive strength by utiliz-
ing the equation derived from Ryshkewitch’s studies [20].
The fitted equation is as mentioned below:
𝜎 = 𝜎0 × e−B×P (1)
where σ is the strength of porous body, σ0 is the strength of
nonporous body of the same concrete and P is porosity of
the concrete.
In this equation, σ0 and B are arbitrary constants. Equa-
tion (1) mentioned above indicates an exponential curve
which was obtained by varying the strength along with AP
and MP for all concrete samples. The obtained values of
arbitrary constants are shown in Table 3. Besides the data
provided the R-squared value of each curve fitting was
almost similar and approximately 0.67.
Utilizing the procured values of corresponding arbi-
trary constants, the curves of variation of AP and MP with
strength were plotted for Type 1 and Type 2 compaction
distributions, separately. Figure 3 shows the variation of AP
and MP drawn against strength for all pervious concrete mix
designs of type 01 and type 02.
The graph well illustrates that the AP is higher than MP
for a fixed strength for both types of compaction distribution.
Moreover, it is evident that both AP and MP show a gradual
decrement along with the increment of strength. Based on
our available data, strength of the pervious concrete cubes
and MP increase with A/C ratio for a particular compaction
of both types.
While observing the compressive strengths of Type
01 and type 02 compaction distribution in mix designs,

Fig. 3  Variations of Strength

with porosity (AP and MP) for
Type 1 and Type 2 mix designs

13
Investigation of Compaction on Compressive Strength and Porosity of Pervious Concrete
irrespective of porosity values, the strength of type 01 was
higher than that of type 02 for all the concrete cubes. Hence,
it could be concluded that compaction type 01 is ideal to
produce pervious concrete with high porosity values.
In addition, when the correlation of MP and AP values
were analysed, there was no definite distinction between the
types of compaction in the correlation between MP and AP
as evident from Fig. 4. Nevertheless, MP increased with
the increment of AP. Thereby it is perceptible that both MP
and AP values affect the performance of pervious concrete.
It is noticeable from the linear relation between MP and
AP that the slope was not equal to one or a value closer to
one. In addition, AP measures the total porosity, while MP
measures only the accessible fraction of porosity. In other
words, MP would be subjected to the availability of pores
that are exposed to the surface, which may vary depending
on the porosity of the cube (lower porosity may have more
disconnected pores, while higher porosity samples may have
higher degree of connections between the pores and to the
surface). Since AP has almost sufficient correlation with
compressive strength, we could infer that MP is not an ideal
indicator for comparison of properties of pervious concrete
for various mix designs. Subsequently AP is comparatively a
better indicator for accurate comparison of properties related
to pervious concrete.
4.2 Analysis of Effective Porosity (EP) Against
Actual Porosity (AP)
Furthermore, to analyze the distribution of porosity along
the vertical profile the image obtained was then split into
10 layers horizontally (the thickness of each layer was 15
mm, whereas the total height of the image or the height
of core sample was 150 mm) and porosity of each layer
was computed. Figure 5 shows the variation of porosity
Fig. 4  MP vs AP for all the con-
cretes in Type 1 mix designs
with depth from top and A/C ratios for pervious concrete
samples cast with 30 blows of compaction.
From Fig. 5, it could be observed that a layer on both
ends (top and bottom of the cube) had significantly
higher porosity than the rest of the volume of cube which
accounts for all cubes as it was observed in previous stud-
ies [49]. However, throughout the inner layers, distribution
of porosity was mostly uniform for all pervious concrete
cube samples, irrespective of compaction distribution or
mix designs. The term Effective porosity is defined, there-
fore, as the mean porosity of the inner layers of the cube,
accounting for middle eight layers. As EP is a value repre-
senting approximately 80% of the sample, it is expected to
serve as an indicator for mechanical properties of pervious
concrete.
For in-depth analysis of data, SPSS, a powerful statisti-
cal software was used. First, Multivariate Analysis of Vari-
ance (MANOVA) was done for each type of compaction
separately and analysed. MANOVA was chosen to be the
statistical analysis employed in this study as there was a need
to examine the effects of more than one independent vari-
ables on multiple dependent variables. Dependent variables
considered for this analysis are Strength, EP, AP, whereas
the independent variables are A/C ratio and blows. In mul-
tivariate test, tests of between-subject effects were utilized
to analyze the significant level of dependent variable with
Mean square and F value. Tables 4 and 5 show the results
obtained from MANOVA test for type one and type two
compaction, respectively, to show the overall summary of
dependent variable with independent variable.
Here the product of A/C ratio and number of blows given
were taken as the independent factor. The intercept row
depicts the hypothesis matrix for testing the significance of
dependent variable in within-subjects factor. The Corrected
Model values in this table provide the data regarding total
13
 M. Sajeevan et al.

Fig. 5  Porosity vs depth graph

for all aggregate-to-cement
ratio designs with 30 blows
compaction

Table 4  Tabulation of MANOVA test results for both the type 01 Table 5  Tabulation of MANOVA test results for both the type 02
compaction compaction
Source Dependent variable F Sig Source Dependent Variable F Sig

Corrected model Strength 111.031 0.000 Corrected Model Strength 361.391 0.000
Actual porosity 51.276 0.000 Actual porosity 31.213 0.000
Effective porosity 44.506 0.000 Effective porosity 35.018 0.000
Intercept Strength 8389.574 0.000 Intercept Strength 30,519.512 0.000
Actual porosity 12,845.458 0.000 Actual porosity 6028.062 0.000
Effective porosity 8083.471 0.000 Effective porosity 5339.308 0.000
A/C ratio Strength 274.764 0.000 A/C ratio Strength 623.806 0.000
Actual porosity 130.880 0.000 Actual porosity 85.009 0.000
Effective porosity 115.513 0.000 Effective porosity 94.071 0.000
Blows Strength 124.826 0.000 Blows Strength 800.636 0.000
Actual porosity 58.429 0.000 Actual porosity 28.416 0.000
Effective porosity 45.211 0.000 Effective porosity 34.328 0.000
A/C ratio * Blows Strength 19.024 0.000 A/C ratio * Blows Strength 81.685 0.000
Actual porosity 5.109 0.000 Actual porosity 2.185 0.000
Effective porosity 4.264 0.000 Effective porosity 2.495 0.000

sum of A/C ratio, number of blows given and the product
of both.
As part of performing a MANOVA on this data F-ratio
was computed which is defined as the ratio between mean
square of factors and means square of error. The calculated
F-ratio can be compared to a table of critical F-ratios to
determine if there are any differences between groups or
not. The Sig. column determines whether the multivariate
test was statistically significant or not for 95% confidence,
which is indicated by a value less than 0.05.
From Table 4, it is well-established that throughout the
analysis the value corresponding to significant value was
less than 0.001 in all the cases. Moreover, while noting
the data related to intercept, it clearly mentions that each
factor induced different impacts on dependent variables.
Significant level pertaining to all other cases was also zero
which in turn strengthens the fact that A/C ratio and number
of blows given affects the strength and porosity of pervious
concrete. Similar observations were evident from the data
of type two compaction shown in Table 5.
Post hoc test was performed to determine which specific
independent variable level significantly differs from another
which is to compare the data within each pair of A/C ratio
and number of blows given. Tukey’s HSD, a post hoc test
used to find out which specific groups’ means are differ-
ent from each factor. Pairs of groups which does not have
significant difference among them based on each factor and
dependent variable are tabulated in Table 6.
The table gives a clear view of data regarding A/C ratio
and number of blows given for both the compaction types

13
Investigation of Compaction on Compressive Strength and Porosity of Pervious Concrete

Table 6  Post hoc test results for both the compaction types When AP is considered the A/C ratio changes with the
Type A/C ratio or blows Dependent variable
compaction given. When a lower A/C ratio is employed, the
binder content would be squeezed out during the compaction
1 A/C ratio Strength 6.0–7.0 process. Therefore, the actual A/C ratio will be higher than
Actual porosity 3.5–4.0 that of the initial mix design. When a higher A/C ratio is
4.0–5.0
employed the binder content will penetrate through the pores
4.5–5.0
6.0–7.0 during the process of compaction, and therefore, the actual
Effective porosity 3.5–4.0 A/C ratio will be lesser than that of the initial mix design.
4.5–5.0 Both the above-mentioned conditions create similarity with
6.0–7.0 nearest group.
Blows Strength – The corresponding values of strength with number of
Actual porosity 45–60 blows given for both the compaction types differ from each
60–75
other. However, when AP and EP are taken as dependent
Effective porosity 45–60
variables under the number of blows, there is no difference
60–75
in the values. This is due to the fact that above 45 blows, it
2 A/C ratio Strength 6.0–7.0
has less or no efficiency. Figure 6 adds more reference to the
Actual porosity 3.5–4.0
4.5–5.0 statement. Figure 6 shows the error bar graph for the vari-
4.5–7.0 ation of AP and EP with the number of blows for a specific
5.0–7.0 A/C ratio of 2.5 with number of blows given.
6.0–7.0
Although there are certain fluctuations in the porosity val-
Effective porosity 3.5–4.0
ues for compaction with number of blows less than 45, there
4.5–5.0
4.5–7.0 is no distinctive variation in porosity values as we increase
5.0–6.0 the number of blows above 45. Moreover, the porosity values
5.0–7.0 are comparatively lesser as we move along the number of
6.0–7.0
blows above 45. For all mix designs, AP was greater than
Blows Strength –
EP as presumed. Figure 7 shows the correlation between AP
Actual porosity 45–60
and EP for both compaction types.
60–75
When AP and EP are plotted to a linear correlation, the
Effective porosity 45–60
60–75 value of ­R2 obtained was closer to one which indicates a
very high correlation of the model to the data. The gradient
value of the curve is also observed to be of a value closer
separately. While focusing on the A/C ratio, for both the to one (0.9701). Moreover, AP was approximately equiva-
compaction types the value corresponding to strength is lent to (EP + 3.5%) when considering all mix designs. The
6.0–7.0. This is because when increasing the A/C ratio the deviation between AP and EP was independent of the total
binder content will reduce vastly. Thus, bonding among the number blows given and the compaction type. AP and EP
particles would be quite less for both A/C ratios 6.0 and 7.0. could be, therefore, considered as surrogate parameters, even

Fig. 6  Error bar graph for the

variation of AP and EP with the
number of blows for a specific
A/C ratio of 2.5 with number
of blows

13

Fig. 7  EP vs AP for all pervious
concretes used in this experi-
ment
though the absolute values may differ. The variation of AP
and EP was further analysed for compaction types 1 and 2
separately, as shown in Figs. 8 and 9, respectively.
Although the values of R2 was similar for both types of
compaction, while comparing the equations of respective
curves with each other, there was a significant difference
between gradient coefficients. For example, the gradient
of the curve for Type 1 indicates almost a perfect match
between AP and EP (gradient of 0.99), while for Type 2,
a difference could be observed as indicated by a lesser
gradient value (0.94). Although the slope of both curves
corresponding to type 01 and 02 approximately similar,
there was a distinguishable difference within the cut-off
point values of both the equations (4.00 and 2.8 for type
01 and type 02, respectively). The reason for this devia-
tion was probably due to the distribution of compaction
energy of concrete cubes throughout the vertical profile.
Fig. 8  EP vs AP for all the con-
cretes in Type 1 mix design
13
M. Sajeevan et al.
Furthermore, the variation of EP and AP with A/C ratio
for a fixed value of 45 blows was analysed for both the
compaction types, as shown in Fig. 10.
The porosity values (EP an AP) gradually increased with
the increment of A/C ratio irrespective of the type of com-
paction. In addition, a T test was also performed to test how
significant the differences were between both the types.
Paired T test was utilized in this study to analyze how sig-
nificant the difference is in between both compaction types
for each of AP and EP data. Table 7 shows the results of the
paired T test carried out, which showed the existence of a
significant difference between Type 1 and Type 2 compac-
tion for both EP and AP. The value of Pearson’s R, the cor-
relation coefficient, is approximately similar for both AP and
EP values. Moreover, the values are much closer to positive
one which in turn indicates that it is almost a perfect posi-
tive correlation.
Investigation of Compaction on Compressive Strength and Porosity of Pervious Concrete

Fig. 9  EP vs AP for all the con-

cretes in Type 2 mix design

Fig. 10  EP and AP with A/C

ratio for a fixed value of 45
blows for both the compaction
types

Table 7  Results obtained from paired T test done for AP and EP sep- the classes each case represented based on classification, to
arately for both the compaction types compute the accuracy of classification.
N Pearson Cor- Sig Machine learning tools were then employed to find rela-
relation tionship between the functions of both types which showed
AP Type 1 and Type 2 240 0.949 0.000 variation of EP with AP. 80% of the entire data were taken
into account for model development (training) and the
EP Type 1 and Type 2 240 0.960 0.000
remaining 20% used for validation of the model (testing).
The k-fold cross validation was followed in this study, where
k was equal to ten that partitioned the data into equally sized
4.3 Analysis of EP and AP Against Type 01 and Type ten folds. The algorithms applied in this data problem and
02 Compactions, Using ML Tools their corresponding accuracies are tabulated in Table 8.
The accuracy values indicated by the percentage of cor-
The relationship between EP and compressive strength of rect predictions for the test data are higher for certain algo-
pervious concrete was analysed using Machine Learning rithms that includes Decision Tree (DT), Random Forest
(ML) classification tools. The cases were classified into (RF) and KNN Classifier. Thereby, these were considered
two classes using 09 different classification ML tools, and as selected classifiers for further analysis and an accuracy
assessed if the classification of cases represented type of of model prediction was computed, as shown in Table 9.
compaction used (Type 01 and Type 02). The classifica- From the above discussion we could conclude that EP
tions were then tested against the predefined classes and and AP imparted similar impact on performance of pervious

13
 M. Sajeevan et al.

Table 8  Results obtained from nine distinct algorithms of machine vertical profile of core samples of pervious concrete cubes,
learning as shown in Fig. 5. Although compaction is given in three
Classifier Accuracy layers in this experiments, the distribution of compaction
energy along the vertical and horizontal profile may not be
Naive bayes 0.39
uniform. It is essential to analyse the impact of compac-
Logistic regression 0.48
tion on the packing of aggregates and pores in pervious
Support vector machine—linear kernel 0.5
concrete, in the vertical profile. Hence, the variation of
Support vector machine—RBF kernel 0.5
porosity differences with respect to the preceding layer is
Support vector machine—polynomial kernel 0.48
computed to understand the variation along the vertical
Decision tree 0.64
profile within the pervious concrete cube samples.
Random forest 0.73
Figure 11 shows the variation of porosity difference
KNN classifier 0.66
(as a percentage of change with respect to the preceding
Perceptron 0.5
strip porosity value) within layers drawn against the depth
of core sample of pervious concrete cubes cast with 45
blows of compaction for different A/C ratios. Each point
Table 9  Selected classifiers and their respective accuracy models in the graph depicts the value of increment or decrement of
Selected classifier Accuracy model
porosity with respect to its preceding slice. It is apparent
from Fig. 11 that the distribution of differences in poros-
Decision tree 0.65 ± 0.11 ity values takes a shape of an ‘S’ curve. The variation of
Random forest 0.65 ± 0.092 porosity difference values at both ends of the curve shows
KNN classifier 0.61 ± 0.087 a deviation for various A/C ratios. The points represented
in the curve does not follow any gradual increments with
respect to design parameters and has step-ups and step-
concrete. Therefore, both EP and AP can be used quantita- downs in porosity with depth. However, a similar pattern
tively analyse the performance of pervious concrete. Owing observed in all mix designs pertaining to different compac-
to the fact that the evaluation of AP values is comparatively tion levels as observed in the figure.
easier than that of EP, AP is considered to be the most appro- The coefficient of determination value, ‘R 2’ is almost
priate and accurate indicator of porosity in pervious concrete above 0.6 for all the curves which indicated that third-
for comparison of performance. In addition, the uncertainty order polynomial is a best fit for the curve. A third-order
of porosity values observed was within 7% of the mean polynomial curve could be represented, as shown in the
across all designs. following equation:

4.4 Porosity Distribution in the Vertical Profile y = ax3 + bx2 + cx + d (2)

where a, b, c and d in equation, are arbitrary constants,

On the other hand, there are certain fluctuations in the dis- which were obtained for all samples. However, there is not
tribution of actual porosity values as we moved along the any distinctive relationship observed in the equation of the
curves corresponding to their respective A/C ratios.

Fig. 11  Difference in poros-

ity along the vertical profile vs
depth graph for all aggregate-
to-cement ratio designs with 45
blows compaction

13
Investigation of Compaction on Compressive Strength and Porosity of Pervious Concrete
Similar to Fig. 11, the variation of differences in porosity
with depth is analysed by varying the compaction type and
the number of total blows given. This is depicted in Fig. 12.
In this figure, the variation of differences in porosity dis-
tribution is plotted for pervious concrete cubes casted with
compaction types 01 and 02 for 15, 30 and 45 blows for an
A/C ratio of 2.5. Similar to Fig. 11, this curve also takes up
the shape of ‘S’. There is no linear relationship present with
the increase in number of blows. However, the value of ­R2
being greater than 0.6 endorses the fact that the variation fits
best to third-order polynomial of equation, Fig. 12.
It is apparent that both A/C and compaction have dif-
ferent models and that cannot be compared with absolute
values. Thus, a standardized model needed to be drawn to
compare observations across different designs of pervious
concrete. Z-scores are used to standardize the data sets and
enable comparison pattern between different populations
of data. Z-score allows comparison of observations from
mean, when measured in standard deviation units (Wang
et al., 2017). Figure 13 depicts the variation of Z-scores
of porosity difference within layers with depth for all A/C
Fig. 12  Difference in poros-
ity along the vertical profile
vs depth graph for compaction
types 01 and 02 corresponding
to 15, 30 and 45 blows
Fig. 13  Z-score values of
porosity differences of cubes
for all A/C ratios and 15 blows
of compaction in type 02 with
depth
ratios for concrete cube samples casted for 15 Blows of
compaction Type 02. As the values are standardized, data
sets with different standard deviation and mean can be
compared irrespective of the porosity parameters.
The value of R2 being greater than 0.7 which is a closer
value to 1.0, emphasizes that each data point lies close to
the predicted values. Moreover, the variation of Z-scores
of porosity difference within layers for concrete cube sam-
ples cast with all compactions and all A/C ratios for both
compaction Type 01 and 02 against depth is plotted, as
shown in Fig. 14.
Figure 14 shows that the compaction distribution does
not affect porosity distribution in the vertical profile.
The values of arbitrary constants of curve fitting and ­R2
for Type 01, Type 02, and Type 01 + 02 are tabulated in
Tables 8 and 10.
The coefficients tabulated in Table 8, clearly indicates
that there is no difference between both types and that the
compaction types had no significance impact on the distribu-
tion of porosity in the vertical profile. Therefore, the func-
tion of the Z score value can be given, as shown in Eq. 3.
13
 M. Sajeevan et al.

Fig. 14  Z-score values of poros-

ity differences of cubes for all
A/C ratios and all compaction
designs in different compaction
distributions with depth

Table 10  Values of coefficients a, b, c, d corresponding to equation distribution is observed to be effective in compacting

compaction for types
Type 2 Type 1 Type 1 + 2
a 1.00E-05 1.00E-05 1.00E-05
b − 0.0029 − 0.0028 − 0.0028
c 0.1903 0.1896 0.19
d − 3.7795 − 3.7856 − 3.7826
R2 0.658 0.6707 0.6643
The value of the coefficient of determination obtained from
both compaction types are around 0.7:
y = 1 × 10−5 x3 − 0.0028x2 + 0.19x − 3.7826
5 Conclusion
Most significant property of pervious concrete is poros-
ity, which in turn influences the strength of the materials
in construction. The current study analysed the impact of
compaction on the performance of pervious concrete, and
the following conclusions are made;
1. The widely adopted Ryshkewitch’s model for porosity
and compressive strength yielded an ­R2 of 0.67 in this
study (when all designs are considered and also for dif-
ferent types of compaction, separately). However, a sig-
nificant difference in the constant σ0 has been observed
(50.17 and 40.13 against 29.9 and 23.59 for AP and MP
for type 01 compaction: equally distributed compaction
in three layers and type 02 compaction: different distri-
butions of compaction in three layers, respectively).
2. Considering the compressive strength, type 01 compac-
tion (equally distributed compaction in three layers)
the matrix of pervious concrete. The porosity on the
other hand did not get affected by the type of compac-
tion employed, that ensured that the optimum compres-
sive strength was achieved without compromising on the
porosity.
3. The relationship between MP and AP is unaffected by
the type of compaction employed. Owing to the differ-
ence between MP and AP, however, AP is considered to
represent the porosity of pervious concrete.
4. While MANOVA test returned existence of significant
impact of A/C ratio and both type of compaction on
compressive strength and porosity, post hoc test showed
this impact to have different impacts in the different
(3) ranges of mix design parameters. In general, the impact
of compaction (number of blows) on porosity is not sig-
nificant between 60 and 75 for both types of compaction.
Similar observation was made on the impact of A/C ratio
on AP and EP and that it was consistent for both types
of compaction [50–52].
5. The relationship between AP and EP is observed to dif-
fer between the types of compaction. Machine Learn-
ing methods of classification, such as Random Forest
(RF), Decision Tree (DT) and KNN Classifier (KNN)
yielded an accuracy of 0.73, 0.64 and 0.66, respectively,
in classifying the relationships based on type of com-
paction. The accuracy remained within an error of less
than 8% in a tenfold cross validation, indicating a strong
and distinct difference between the type of compaction
employed.
Funding National Research Council Sri Lanka, 19-045, Daniel Niru-
ban Subramaniam

13
Investigation of Compaction on Compressive Strength and Porosity of Pervious Concrete
Declarations
Conflict of Interest The authors declare that they have no conflict of
interest.
References
1. Alvarez-Larrauri, R., & Fogel, I. (2008). Environmental audits
as a policy of state: 10 years of experience in Mexico. Journal of
Cleaner Production, 16, 66–74.
2. Anburuvel, A., Subramaniam, D. N., 2022b. Investigation of the
Effects of Compaction on Compressive Strength and Porosity
Characteristics of Pervious Concrete. Transportation Research
Record, 0, 03611981221087236.
3. Subramaniam, DN., Sathiparan, N. 2022. Comparative study
of fly ash and rice husk ash as cement replacement in pervious
concrete: mechanical characteristics and sustainability analysis.
International Journal of Pavement Engineering, 1–18.
4. Yang, J., & Jiang, G. (2003). Experimental study on properties
of pervious concrete pavement materials. Cement and Concrete
Research, 33, 381–386.
5. Pal, M. (2005). Random forest classifier for remote sensing clas-
sification. International Journal of Remote Sensing, 26, 217–222.
6. Shen, W., Shan, L., Zhang, T., Ma, H., Cai, Z., & Shi, H. (2013).
Investigation on polymer–rubber aggregate modified porous con-
crete. Construction and Building Materials, 38, 667–674.
7. Yahia, A., & Kabagire, K. D. (2014). New approach to propor-
tion pervious concrete. Construction and Building Materials, 62,
38–46.
8. Ring, M., & Eskofier, B. M. (2016). An approximation of the
gaussian RBF kernel for efficient classification with SVMs. Pat-
tern Recognition Letters, 84, 107–113.
9. Adabre, M. A., Chan, A. P. C., Darko, A., Osei-Kyei, R., Abi-
doye, R., & Adjei-Kumi, T. (2020). Critical barriers to sustain-
ability attainment in affordable housing: international construction
professionals’ perspective. Journal of Cleaner Production, 253,
119995.
10. Kononenko, I. Semi-naive bayesian classifier. In: Kodratoff, Y.,
eds. Machine Learning — EWSL-91, 1991// 1991 Berlin, Heidel-
berg. Springer Berlin Heidelberg, 206–219.
11. Brodie, I. (2007). Prediction of stormwater particle loads from
impervios urban surfaces based on a rainfall detachment index
water science and technology. Journal Of The International Asso-
ciation on Water Pollution Research, 55, 49–56.
12. Ahilash, N., Sajeevan, M., Subramaniam, D. N. Optimising pervi-
ous concrete design with partial replacement of cement with fly
ash. 12th International Conference on road and airfield Pavement
Technology, 2021 Sri Lanka.
13. Bhutta, A., Abdul Shukor Lim, N. H., Ariffin, N. F., Hussin, M.,
Tahir, M., & Mirza, J. (2013). Properties of porous concrete from
waste crushed concrete (recycled aggregate). Construction and
Building Materials, 47, 1243–1248.
14. Brown, R. A., & Borst, M. (2015). Nutrient infiltrate concentra-
tions from three permeable pavement types. Journal of Environ-
mental Management, 164, 74–85.
15. Carmichael, M. J., Arulraj, G. P., & Meyyappan, P. L. (2021).
Effect of partial replacement of cement with nano fly ash on per-
meable concrete: a strength study. Materials Today: Proceedings,
43, 2109–2116.
16. Djerbi, A. (2018). Effect of recycled coarse aggregate on the new
interfacial transition zone concrete. Construction and Building
Materials, 190, 1023–1033.
17. Carsana, M., Tittarelli, F., & Bertolini, L. (2013). Use of no-fines
concrete as a building material: Strength, durability properties
and corrosion protection of embedded steel. Cement and Concrete
Research, 48, 64–73.
18. Subramaniam, D. N., Logeswaran, T., Tharshikka, V., & Nilak-
shan, B. (2018). Dynamics of clay particles in non-vegetated
stormwater biofilters. Water, Air, & Soil Pollution, 229, 302.
19. Anburuvel, A., Subramaniam, D. N., (2022a). Influence of
aggregate gradation and compaction on compressive strength
and porosity characteristics of pervious concrete. International
Journal of Pavement Engineering, 1–14.
20. Ryshkewitch, E. (1953). Compression strength of porous sintered
alumina and zirconia. Journal of the American Ceramic Society,
36(2), 65–8.
21. Ba, M.-F., Qian, C.-X., Guo, X.-J., & Han, X.-Y. (2011). Effects
of steam curing on strength and porous structure of concrete with
low water/binder ratio. Construction and Building Materials, 25,
123–128.
22. Lyu, K., Garboczi, E. J., She, W., & Miao, C. (2019). The effect
of rough vs. smooth aggregate surfaces on the characteristics of
the interfacial transition zone. Cement and Concrete Composites,
99, 49–61.
23. Ramjeawon, T. 2020. Introduction to Sustainability for Engineers.
24. Carleo, G., Cirac, I., Cranmer, K., Daudet, L., Schuld, M., Tishby,
N., Vogt-Maranto, L., & Zdeborová, L. (2019). Machine learn-
ing and the physical sciences. Reviews of Modern Physics, 91,
045002.
25. EL Naqa, I., Murphy, MJ., 2015. What Is Machine Learning? In:
EL NAQA, I., LI, R. & MURPHY, M. J. (eds.) Machine Learning
in Radiation Oncology: Theory and Applications. Cham: Springer
International Publishing.
26. Debnath, B., & Sarkar, P. (2018). Pervious concrete as an alterna-
tive pavement strategy: A state-of-the-art review. International
Journal of Pavement Engineering, 21, 1–16.
27. Khan, M.I. (2002). Factors affecting the thermal properties of
concrete and applicability of its prediction models. Building and
Environment 37(6), 607–614.
28. Deo, O., & Neithalath, N. (2011). Compressive response of pervi-
ous concretes proportioned for desired porosities. Construction
and Building Materials, 25, 4181–4189.
29. Ćosić, K., Korat, L., Ducman, V., & Netinger, I. (2015). Influence
of aggregate type and size on properties of pervious concrete.
Construction and Building Materials, 78, 69–76.
30. Islam, M. J., Wu, Q. M. J., Ahmadi, M., SID-Ahmed, MA., Inves-
tigating the performance of naive- bayes classifiers and k- nearest
neighbor classifiers. 2007 international Conference on conver-
gence information technology (ICCIT 2007), 21–23 Nov. 2007
2007. 1541–1546.
31. Zhong, R., & Wille, K. (2016). Linking pore system character-
istics to the compressive behavior of pervious concrete. Cement
and Concrete Composites, 70, 130–138.
32. Jain, A. K., Chouhan, D. J. S. 2011. Effect of shape of aggregate
on compressive, strength and permeability properties of pervious
concrete.
33. Kazemi, F., & Hill, K. (2015). Effect of permeable pavement
basecourse aggregates on stormwater quality for irrigation reuse.
Ecological Engineering, 77, 189–195.
34. Agar Ozbek, A. S., Pedersen, R. R., Weerheijm, J., & Van Breugel,
K. (2019). Mesoscopic modeling of the impact behavior and frag-
mentation of porous concrete. Cement and Concrete Composites,
102, 116–133.
35. Chandrappa, A. K., & Biligiri, K. P. (2016). Pervious concrete as
a sustainable pavement material – Research findings and future
prospects: a state-of-the-art review. Construction and Building
Materials, 111, 262–274.
13

36. Alshareedah, O., & Nassiri, S. (2020). Pervious concrete mixture
optimization, physical, and mechanical properties and pavement
design: A review. Journal of Cleaner Production, 288, 125095.
37. Bonicelli, A., Giustozzi, F., & Crispino, M. (2015). Experimental
study on the effects of fine sand addition on differentially com-
pacted pervious concrete. Construction and Building Materials,
91, 102–110.
38. Fisher, R., et al. (1996). Hypermedia image processing reference.
England: John Wiley & Sons Ltd, p. 118–130
39. Jain, A., Metzger, M. J., & Glasser, B. J. (2013). Effect of parti-
cle size distribution on segregation in vibrated systems. Powder
Technology, 237, 543–553.
40. Lavalley, M. P. (2008). Logistic regression. Circulation, 117,
2395–2399.
41. Lucke, T., & Nichols, P. W. B. (2015). The pollution removal
and stormwater reduction performance of street-side bioretention
basins after ten years in operation. Science of The Total Environ-
ment, 536, 784–792.
42. Chou, J.-S., Tsai, C.-F., Pham, A.-D., & Lu, Y.-H. (2014).
Machine learning in concrete strength simulations: Multi-nation
data analytics. Construction and Building Materials, 73, 771–780.
43. Ben Chaabene, W., Flah, M., & Nehdi, M. L. (2020). Machine
learning prediction of mechanical properties of concrete: critical
review. Construction and Building Materials, 260, 119889
44. Chauhan, V. K., Dahiya, K., & Sharma, A. (2019). Problem for-
mulations and solvers in linear SVM: A review. Artificial Intel-
ligence Review, 52, 803–855.
45. Sajeevan, M., Ahilash, N., Subramaniam, D. N. Investigating
impact of boundary layer in pervious concrete. 12th International
Conference on Road and Airfield Pavement Technology, 2021 Sri
Lanka.
46. Boser, B. E., Guyon, I. M., Vapnik, V. N. A training algorithm for
optimal margin classifiers. Proceedings of the fifth annual work-
shop on Computational learning theory, 1992. 144–152.
47. Rinduja, R., Sajeevan, M., Thajani, J., Subramaniam, DN., Inves-
tigation of porosity distribution in pervious cocnrete. 8th Interna-
tional Symposium on advances in Civil and Environmental Engi-
neering Practice for Sustainable Development, 2021 Sri Lanka.
48. Belue, L. M., & Bauer, K. W. (1995). Determining input features
for multilayer perceptrons. Neurocomputing, 7, 111–121.
49. Subramaniam, D. N., Hareindirasarma, S., Janarthanan, B. 2022.
An Alternative Approach to optimize aggregate-to-cement ratio
and compaction in pervious concrete. Arabian Journal for Science
and Engineering.
50. Grubeša, I. N., Barišić, I., Ducman, V., & Korat, L. (2018). Drain-
ing capability of single-sized pervious concrete. Construction and
Building Materials, 169, 252–260.
13
M. Sajeevan et al.
51. Lian, C., Zhuge, Y., & Beecham, S. (2011). The relationship
between porosity and strength for porous concrete. Construction
and Building Materials, 25, 4294–4298.
52. Sajeevan, M., Subramaniam, D. 2022. Investigation of boundary
layer impact on Pervious Concrete. International Journal of Pave-
ment Engineering.
Springer Nature or its licensor (e.g. a society or other partner) holds
exclusive rights to this article under a publishing agreement with the
author(s) or other rightsholder(s); author self-archiving of the accepted
manuscript version of this article is solely governed by the terms of
such publishing agreement and applicable law.
M. Sajeevan is an undergraduate student, currently in his final year of
reading for BSc. Eng. in Civil and Environmental Engineering, from
University of Jaffna. He contemplates on embarking on research and
postgraduate training in sustainable materials after graduation.
D. N. Subramaniam is currently a senior lecturer attached to the
Department of Civil Engineering of University of Jaffna. He has com-
pleted his doctoral degree from Queensland University of Technology,
Australia, and MSc from University of Southampton, UK, on water
treatment systems. He is an environmental engineering, conducting
teaching and research on environmental engineering-related topics,
including Water Sensitive Urban Design. His current research inter-
est lies in Pervious concrete (as a structural element solution for low
impact development). His other research avenues include water treat-
ment engineering, recycled concrete aggregates, solid waste manage-
ment and coastal sediment transport.
R. Rinduja is an undergraduate student of University of Jaffna, cur-
rently reading in third year of BSc. Eng in Civil and Environmental
Engineering. She joined the University Research Group on Pervious
Concrete in her first year and continued to participate in research activi-
ties of the group.
J. Pratheeba is currently serving as a senior lecturer attached to the
Department of Computer Engineering. She obtained her doctoral
degree from University of Southampton, UK, working on a research
on bioinformatics. She is involved in the research group on pervious
concrete, contributing to research activities related to image process-
ing, image analysis and employing machine learning tools for model
development. Her main research lies on bioinformatics and machine
learning techniques.