Professional Documents
Culture Documents
Full Chapter Progress in Advanced Computing and Intelligent Engineering Proceedings of Icacie 2019 Volume 2 Chhabi Rani Panigrahi PDF
Full Chapter Progress in Advanced Computing and Intelligent Engineering Proceedings of Icacie 2019 Volume 2 Chhabi Rani Panigrahi PDF
https://textbookfull.com/product/progress-in-advanced-computing-
and-intelligent-engineering-proceedings-of-
icacie-2019-volume-1-chhabi-rani-panigrahi/
https://textbookfull.com/product/progress-in-advanced-computing-
and-intelligent-engineering-proceedings-of-
icacie-2016-volume-2-1st-edition-khalid-saeed/
https://textbookfull.com/product/progress-in-advanced-computing-
and-intelligent-engineering-proceedings-of-
icacie-2016-volume-1-1st-edition-khalid-saeed/
https://textbookfull.com/product/advanced-computing-and-
intelligent-engineering-proceedings-of-icacie-2018-volume-2-1st-
edition-bibudhendu-pati-editor/
Intelligent Computing: Proceedings of the 2018
Computing Conference, Volume 2 Kohei Arai
https://textbookfull.com/product/intelligent-computing-
proceedings-of-the-2018-computing-conference-volume-2-kohei-arai/
https://textbookfull.com/product/intelligent-computing-
proceedings-of-the-2020-computing-conference-volume-2-kohei-arai/
https://textbookfull.com/product/computing-in-engineering-and-
technology-proceedings-of-iccet-2019-brijesh-iyer/
https://textbookfull.com/product/intelligent-computing-
proceedings-of-the-2020-computing-conference-volume-3-kohei-arai/
Advances in Intelligent Systems and Computing 1199
Progress in
Advanced
Computing
and Intelligent
Engineering
Proceedings of ICACIE 2019, Volume 2
Advances in Intelligent Systems and Computing
Volume 1199
Series Editor
Janusz Kacprzyk, Systems Research Institute, Polish Academy of Sciences,
Warsaw, Poland
Advisory Editors
Nikhil R. Pal, Indian Statistical Institute, Kolkata, India
Rafael Bello Perez, Faculty of Mathematics, Physics and Computing,
Universidad Central de Las Villas, Santa Clara, Cuba
Emilio S. Corchado, University of Salamanca, Salamanca, Spain
Hani Hagras, School of Computer Science and Electronic Engineering,
University of Essex, Colchester, UK
László T. Kóczy, Department of Automation, Széchenyi István University,
Gyor, Hungary
Vladik Kreinovich, Department of Computer Science, University of Texas
at El Paso, El Paso, TX, USA
Chin-Teng Lin, Department of Electrical Engineering, National Chiao
Tung University, Hsinchu, Taiwan
Jie Lu, Faculty of Engineering and Information Technology,
University of Technology Sydney, Sydney, NSW, Australia
Patricia Melin, Graduate Program of Computer Science, Tijuana Institute
of Technology, Tijuana, Mexico
Nadia Nedjah, Department of Electronics Engineering, University of Rio de Janeiro,
Rio de Janeiro, Brazil
Ngoc Thanh Nguyen , Faculty of Computer Science and Management,
Wrocław University of Technology, Wrocław, Poland
Jun Wang, Department of Mechanical and Automation Engineering,
The Chinese University of Hong Kong, Shatin, Hong Kong
The series “Advances in Intelligent Systems and Computing” contains publications
on theory, applications, and design methods of Intelligent Systems and Intelligent
Computing. Virtually all disciplines such as engineering, natural sciences, computer
and information science, ICT, economics, business, e-commerce, environment,
healthcare, life science are covered. The list of topics spans all the areas of modern
intelligent systems and computing such as: computational intelligence, soft comput-
ing including neural networks, fuzzy systems, evolutionary computing and the fusion
of these paradigms, social intelligence, ambient intelligence, computational neuro-
science, artificial life, virtual worlds and society, cognitive science and systems,
Perception and Vision, DNA and immune based systems, self-organizing and
adaptive systems, e-Learning and teaching, human-centered and human-centric
computing, recommender systems, intelligent control, robotics and mechatronics
including human-machine teaming, knowledge-based paradigms, learning para-
digms, machine ethics, intelligent data analysis, knowledge management, intelligent
agents, intelligent decision making and support, intelligent network security, trust
management, interactive entertainment, Web intelligence and multimedia.
The publications within “Advances in Intelligent Systems and Computing” are
primarily proceedings of important conferences, symposia and congresses. They
cover significant recent developments in the field, both of a foundational and
applicable character. An important characteristic feature of the series is the short
publication time and world-wide distribution. This permits a rapid and broad
dissemination of research results.
Indexed by SCOPUS, DBLP, EI Compendex, INSPEC, WTI Frankfurt eG,
zbMATH, Japanese Science and Technology Agency (JST), SCImago.
Kuan-Ching Li
Editors
Progress in Advanced
Computing and Intelligent
Engineering
Proceedings of ICACIE 2019, Volume 2
123
Editors
Chhabi Rani Panigrahi Bibudhendu Pati
Department of Computer Science Department of Computer Science
Rama Devi Women’s University Rama Devi Women’s University
Bhubaneswar, India Bhubaneswar, India
This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721,
Singapore
Preface
This volume contains the papers presented at the 4th International Conference on
Advanced Computing and Intelligent Engineering (ICACIE) 2019: The 4th
International Conference on Advanced Computing and Intelligent Engineering
(www.icacie.com) held during 21–23rd December 2019, at Rama Devi Women’s
University, Bhubaneswar, India. There were 284 submissions and each qualified
submission was reviewed by a minimum of two Technical Program Committee
members using the criteria of relevance, originality, technical quality, and presen-
tation. The committee accepted 86 full papers for oral presentation at the conference
and the overall acceptance rate is 29%.
ICACIE 2019, was an initiative taken by the organizers which focuses on
research and applications on topics of advanced computing and intelligent engi-
neering. The focus was also to present state-of-the-art scientific results, to dis-
seminate modern technologies, and to promote collaborative research in the field of
advanced computing and intelligent engineering. Researchers presented their work
in the conference and had an excellent opportunity to interact with eminent pro-
fessors, scientists, and scholars in their area of research. All participants were
benefitted from discussions that facilitated the emergence of innovative ideas and
approaches. Many distinguished professors, well-known scholars, industry leaders,
and young researchers were participated in making ICACIE 2019, an immense
success. We had also an industry panel discussion and we invited people from
software industries like TCS, Infosys, Cognizant, and entrepreneurs.
We thank all the Technical Program Committee members and all reviewers/
sub-reviewers for their timely and thorough participation during the review process.
We express our sincere gratitude to Prof. Padmaja Mishra, Honourable Vice
Chancellor and Chief Patron of ICACIE 2019, to allow us to organize ICACIE
2019, on the campus and for her unending timely support towards organization of
this conference. We would like to extend our sincere thanks to Prof. Bibudhendu
Pati and Dr. Hemant Kumar Rath, General chairs of ICACIE 2019, for their
valuable guidance during review of papers, as well as other aspects of the con-
ference. We appreciate the time and efforts put in by the members of the local
organizing team at Rama Devi Women’s University, Bhubaneswar, India,
v
vi Preface
The book focuses on theory, practice and applications in the broad areas of
advanced computing techniques and intelligent engineering. This two volumes
book includes 86 scholarly articles, which have been accepted for presentation from
287 submissions in the 5th International Conference on Advanced Computing and
Intelligent Engineering held at Rama Devi Women’s University, Bhubaneswar,
India during 21–23rd December, 2019. The first volume of this book consists of 40
numbers of papers and volume 2 contains 46 papers with a total of 86 papers. This
book brings together academic scientists, professors, research scholars and students
to share and disseminate their knowledge and scientific research works related to
advance computing and intelligent engineering. It helps to provide a platform to the
young researchers to find the practical challenges encountered in these areas of
research and the solutions adopted. The book helps to disseminate the knowledge
about some innovative and active research directions in the field of advanced
computing techniques and intelligent engineering, along with some current issues
and applications of related topics.
vii
Contents
ix
x Contents
Dr. Bibudhendu Pati is Associate Professor and Head of the P.G. Department of
Computer Science at Rama Devi Women’s University, Bhubaneswar, India. He
completed his Ph.D. from IIT Kharagpur. Dr. Pati has 21 years of experience in
teaching, research. His interest areas include Wireless Sensor Networks, Cloud
Computing, Big Data, Internet of Things, and Network Virtualization. He has got
several papers published in journals, conference proceedings and books of inter-
national repute. He is a Life Member of Indian Society of Technical Education
(ISTE), Life Member of Computer Society of India, and Senior Member of IEEE.
xv
xvi About the Editors
Fellow of AAAS. Dr. Mohapatra’s research interests are in the areas of wireless
networks, mobile communications, cyber security, and Internet protocols. He has
published more than 350 papers in reputed conferences and journals on these topics.
Dr. Mohapatra’s research has been funded through grants from the National Science
Foundation, US Department of Defense, US Army Research Labs, Intel
Corporation, Siemens, Panasonic Technologies, Hewlett Packard, Raytheon, and
EMC Corporation.
Namrata P. Mohanty(B) , Sweta Shree Dash, Sandeep Sobhan, and Tripti Swarnkar
1 Introduction
Depression is becoming one of the most widely spreading disability causing disorders
around the globe which is expanding at a very fast pace taking more and more subjects
under its paw. It can be caused by various circumstances such as—peer pressure, any
acute disease, family issues, career tensions, etc. Mostly, it is due to changes in brain
waves and the formation of seizures due to persistent feeling of stress and sadness [3].
One of the most effective ways of detecting it is by recording the EEG signals. EEG
signals are noninvasive and low-cost ways of measuring the brain’s electrical activity,
which detects any abnormalities or deviations occurring from the normal brain waves,
thereby helpful in detecting depressive symptoms in the patient. In today’s world the
developing Human–computer interaction, i.e., HCI has made it much more successful
to detect such a complicated disease rather we can say the terrible disability causing
disorder, i.e., Major Depressive Disorder (MDD) with its machine learning techniques.
The main objective of the empirical study is comparing the performance of the two
benchmark classifiers in the classification process of depression using the EEG signals,
such that it can be helpful in predicting depression for the doctors, and thereby letting
them provide the best preventive measures before the onset of depression to the patients.
In this paper, we have performed the experiment by taking the dataset and feeding it into
two most efficient classifiers, i.e., k-NN and ANN. Here, we have successfully classified
depressed patients and normal subjects with an accuracy of 85%.
2 Literature Review
In the year 2008, Brahmi et al. [8], had performed the classification of the EEG signals
using the back-propagation neural network achieving an accuracy and specificity of 93%
and 94%,respectively. In this paper they have tried to distinguish among awake stage
1+REM, stage2, and Slow Wave Stage (SWS) on EEG signal using machine learning
techniques of Neural networks and wavelet packet coefficients. In the year 2011, Hos-
seinifard et al. [10], performed Linear and nonlinear features extraction along with the
classification using the k-NN, LDA, and LR classifiers. In their experiment, they had
obtained an accuracy of 90% classifying depressed patients and normal subjects. Liao
et al. [11] in the year 2017, has carried out the classification of depression using one
of the standard and most efficient classifier SVM to classify depressed patients. In their
experiment, a special robust spectral-spatial EE feature extractor has been used for the
EEG signals to cope up with the absence of biological and psychological markers effi-
ciently. They have obtained an accuracy of 81.23% in their experiment. Chisci et al. [12]
in the year 2010, has carried a seizure prediction model for detecting seizure formation
in the brain which leads to depression, as well as all the associated diseases.
Individuals with depression or anxiety have been bound to experience the ill effects of
epilepsy than those without depression or anxiety. Different cerebrum territories includ-
ing the frontal, temporal, limbic regions are associated with the biological pathogenesis
of depression in individuals with epilepsy [14] [17]. Machine Learning techniques are
of great help for the detection of epilepsy from the analysis of EEG signals [17]. In the
year 2018, Acharya et al. have focused on seizure formation and prediction and basi-
cally how depression is related to seizure formation which is generally due to sudden
change in the electrical activity of the brain [15]. Piotr Mirowski et al. in the year 2009,
have successfully investigated the efficiency of employing bivariate measures to predict
seizures occurring mostly by depression with a sensitivity of 71% [19].
With the advancement of science and technology, machine learning tools, and tech-
niques can easily predict depression from a much earlier time, thereby keeping this dis-
ability causing disorder at the bay [20]. Machine learning algorithm usually learn, extract,
identify, and map underlying pattern to identify groupings of depressed individuals
without constraint [21].
The whole implementation process has been carried out in the system bearing
the following specification: Processor: Intel(R) Pentium (R) CPU N3540 @2.16 GHz
2.16 GHz, Installed Memory (RAM): 8.00 GB, System Type: 64 bit Operating System,
×64-based processor.
Before preprocessing the data, we have to convert it to the .wav form in order to make it
suitable for being used by the Matlab. Here, we have used the edf2wav online browser
[1], for the conversion of the EEG signals to the .wav format. The .wav format is then
imported by the Matlab for data preprocessing and further implementations. The first
step in the preprocessing is the filtering of the raw EEG signals. In this step, the signals
of the specified frequencies in the range of 0–30 Hz containing the alpha, beta, theta,
gamma, and the delta waves get selected and the rest are rejected. The next step is the
Independent Component Analysis (ICA) was performed which helps in the removal of
the artifacts such as the eye blinking, etc., from the selected wave range. Further, the real
values of the data were obtained from the preprocessed EEG signals, which makes it
easier to extract the specific features for the classification process. The obtained dataset
is of size 12.6 MB containing 50 samples having 10240 data points each.
Feature extraction basically refers to identifying any uniquely recognized patterns from
a group of classified data in order to predict its outcomes. These are meant to reduce the
amount of loss of information that has been fed to the system and at the same time, it
simplifies the implementation process due to the reduction in the amount of data. From
the obtained EEG signals it has been observed that physiological features were highly
correlated with the state of arousal among two subjects. A feature can be considered
significant and selected as input to classifier if its absolute correlation is greater for
physiological features among subjects [6].
Selection of highly correlated features helps to exclude less important features affec-
tive state and emotional expressions. Considering the above studies and statistical fea-
tures like the minimum value, the maximum value, the mean value, and the standard
deviation were selected to represent the EEG signals.
6 N. P. Mohanty et al.
3.3 Classification
As we have labeled data so the research work goes under the supervised learning part of
the machine learning.
For our classification, we have taken two widely used classifiers the k-Nearest
Neighbor (k-NN) and Artificial Neural Network (ANN).
edf2wav and the EEGLAB toolbox to get the filtered data, i.e., the EEG signals which
are free from artifacts.
For further data preprocessing, EEGLAB toolbox available online in MATLAB plat-
form have been used, which makes the EEG signals preprocessing a lot more easier. Data
preprocessing over here includes filtering, epoch selection, and independent component
analysis (ICA).
Figure 2 shows the EEGLAB platform we have used for the EEG data preprocessing.
Then the real values are obtained from the filtered signals which are then used for
feature extraction and classification processes. After which, four features were extracted,
i.e., the minimum value, maximum value, mean value, and the standard deviation in order
to obtain better classification results in further processes.
The time taken by the k-NN is nearly 12 h to generate the confusion matrix in our
system while it took nearly 15 h by ANN to compute the results of the 12.6 MB datafile.
In our experiment, the classification accuracies of k-NN and ANN for the train set
are 83.2 and 87.5%. In the training data set, the accuracy of ANN obtained is higher
than that of the k-NN classifier (Table 1 and Fig. 3).
The test set accuracies of both k-NN and ANN also show a similar pattern as seen
from Table 2 and Fig. 4, where ANN possesses higher accuracy than that of the k-NN.
8 N. P. Mohanty et al.
Table 2 Classification accuracies by selected techniques for the test set (Fig. 4)
We have performed the classification process using two famous machine learning
classifiers those are the k-NN and the ANN. Then we have got the confusion matrix for
the analysis of the performance of the two models. In both the cases, we have got the
true positives and the true negatives rate higher for the ANN model. Thus, the accuracy
rate of ANN is nearly 80.3% as compared to that of the k-NN which has an accuracy
of 74.6% as ANN has better processing capacity due to the presence of interconnected
neurons just the same way a human brain does.
Prediction of Depression Using EEG: A Comparative Study 9
Fig. 5. Comparison between the accuracies between train set and test set
Due to the visible differences of accuracies obtained from the two selected tech-
niques, we have plotted a comparison graph so that it will be easier to select a particular
technique for future researches. From Fig. 5, we can notice that ANN gives a better accu-
racy rate than k-NN which is a clear conclusion why more studies should be done on the
neural network that can be of tremendous help in the field of medical and paramedical
sciences.
5 Conclusion
Our work has demonstrated that the neural networks has the potential of predicting
depression with much accuracy than the other machine learning techniques. Though
ANN has given more accurate results than k-NN still the time taken is more in case
of ANN which can be reduced if the dimensionality can be reduced to a greater extent
by selecting more efficient features and also by implementing the model in a system
having higher processor and RAM. Moreover, better research works should be done on
neural networks considering real time data acquisition including complex brain structure
investigation and analysis. Last but not the least depression is something which shouldn’t
be taken lightly and proper check-up by experienced professionals should be done in
due time so as to get rid of this havoc before the onset of its extreme phase.
References
1. European Data Format (EDF). http://www.edfplus.info
2. MathWorks—MATLAB and Simulink for Technical Computing. https://www.mathworks.
com
3. Mallikarjun, H.M., Dr. Suresh, H.N.: Depression level prediction using EEG signals pro-
cessing. In: International Conference on Contemporary Computing and Informatics (IC31),
pp. 928–933 (2014)
10 N. P. Mohanty et al.
21. Dipnall, J.F., Pasco, J.A., Berk, M., Williams, L.J., Dodd, S., Jacka, F.N., Meyer, D.: Why so
GLUMM? Detecting depression clusters through graphing lifestyle-environs using machine-
learning methods (GLUMM). Eur. Psych. 39, 40–50 (2017). https://doi.org/10.1016/j.eurpsy.
2016.06.003
22. Liu, A. et al.: Machine learning aided prediction of family history of depression. In: 2017
New York Scientific Data Summit (NYSDS), New York, NY, pp. 1–4 (2017).https://doi.org/
10.1109/nysds.2017.8085046
23. Sri, K.S., Rajapakse, J.C.: Extracting EEG rhythms using ICA-R. In: IEEE International Joint
Conference on Neural Networks, IJCNN 2008. (IEEE World Congress on Computational
Intelligence), pp. 2133–2138 (2008)
24. Malmivuo, J., Plonsey, R.: Bioelectromagnetism: Principles and Applications of Bioelectric
and Biomagnetic Fields. Oxford University Press (1995)
25. Delorme, A., Makeig, S.: EEGLAB: an open source toolbox for analysis of single trial EEG
dynamics including independent component analysis. J. Neurosci. Methods 134(1), 9–21
(2004)
26. Wu, Y., Ianakiev, K., Govindaraju, V.: Improved k-nearest neighbor classification. Pattern
Recogn. 35(10), 2311–2318 (2002)
27. Ho, C.K., Sasaki, M.: EEG data classification with several mental tasks. In: 2002 IEEE
International Conference on Systems, Man and Cybernetics, vol. 6, p. 4 (2002)
28. About GSIL—Blekinge Institute of Technology—in real life. http://www.bth.se/com/gsil.
Accessed 05 August 2012
Prediction of Stroke Risk Factors for
Better Pre-emptive Healthcare: A
Public-Survey-Based Approach
1 Introduction
We propose to underline the significance of pre-emptive healthcare for cardio-
vascular stroke by discerning behavioural traits which may play a crucial role in
their contribution to the gradual development of health conditions that inclines
to stroke. The behaviours that affect the health in a negative way such as lack
of regular physical activity, lack of calibrated and nutritious food intake, unre-
strained tobacco use and alcohol consumption, etc., if continued for long, most of
the times may result in health conditions that lead to stroke.1 Thus, to prevent
the looming risk of stroke, positive behavioural changes are indispensable.
In the United States (U.S.), stroke is the fifth leading cause of death claiming
one life out of 20 from the total number of deaths per year with more than 0.7
1
https://www.cdc.gov/stroke/behavior.htm.
million stroke patients per year [1]. Moreover, from the government’s perspective,
each year U.S. spends more than $34 billion for stroke which consists of the
cost of personal healthcare services, medicines for stroke treatment and missed
working days2 [1]. For an individual, this underpins the fact that how little
changes towards a healthier lifestyle can eliminate a very significant amount of
expenditure on health.
In the present work, we aim to find out the relation between the behavioural
traits and the chance of stroke using machine learning (ML) techniques and to
further specify which traits dominate the list of possible risk factors regarding
stroke. We apply our analysis on the Behavioral Risk Factor Surveillance System
(BRFSS)3 which is the largest health-related telephonic survey of United States
and contains a significant number of behavioural features. By purely behavioural
features we mean those that are directly controllable or negotiable (or both) by
an individual without any monetary requirement. (excludes insurances). Thus
behaviours influenced by social context, mental health, etc. are excluded for the
most generalization of our result in order to achieve the maximum relevance with
respect to mass awareness thus the less demography-constrained.
To achieve our objective, we first identify the possible risk factors contribut-
ing to stroke from a set of selected behavioural traits by using a GBM (Gradient
Boosting Machine)-based predictive model. Then we venture forward to anal-
yse the impact of the individual features on the model outcome to prove the
soundness of the features statistically.
Our main contributions can be outlined through the following:
2 Related Works
Positive Behavioural Changes in Prevention of Stroke: As discussed in
Sect. 1, most of the times chronic diseases such as stroke are preventable and
their chances of gradual development can be minimized by changing negative
behavioural traits, leading to a healthy lifestyle. Centers for Disease Control
and Prevention (CDC), U.S., argues that a large percentage of stroke cases
can be eliminated by eliminating the three main risk factors: unhealthy diet
2
https://www.cdc.gov/stroke.
3
https://www.cdc.gov/brfss/index.html.
14 D. Banerjee and J. Singh
[24], excessive smoking [2] and lack of enough physical activity4 [26]. Other
epidemiological studies also highlight the cardinal importance of quantifying the
impact of lifestyle on public health due to the possibility of granular level control
by individuals themselves to encounter chronic diseases such as stroke [3].
Machine Learning in Healthcare: Predicting diseases solely on the report of
patient behaviours can be a difficult problem, especially from a survey dataset.
But recent advancement of machine learning algorithms has opened up oppor-
tunities to navigate through this complex problem of disease prediction from
health survey datasets and also to determine risk factors behind the diseases in
contention.
In cases which present the opportunity to impact a particular case of pre-
diction requirement, decision trees are used [4]. A decision tree is composed of
several weak learners [4], which means that the classifier is only slightly or weakly
correlated with the true classification.5 Thus, the performance can be biased in
favour of the majority class of the target in particular cases if the dataset in con-
tention already has much bias or variance or both. Hence, to better the purely
decision-tree-based classifier performance, random forest [5] and gradient boost-
ing [8] may be used. Gradient boosting algorithm [8] uses a gradient descent
procedure which is an iterative method that moves along with the direction of
steepest descent, defined by the negative of the gradient of the function to find
a local minima of that function. Decision trees are used as weak learners in gra-
dient boosting, where they are implemented one at a time unlike in the case
of random forest where this is done all at a time without the gradient descent
procedure that minimizes the loss while adding trees [7]. Overall, random forests
are built to reduce variance [5] whereas gradient boosting reduces bias [8].
Existing Works on Stroke Prediction: Among the existing works on stroke
prediction, Yang, Zhong et al. study the risks in state-level demographics [13].
Akdag et al. implement classification trees for finding the risk factors of hyper-
tension from an observation conducted on hospital patients in Turkey [14]. Sun-
moo Yoon et al. work upon the prediction of disability—one of the results of
stroke—and how the types of disability are associated with stroke and their co-
relation [11]. Alkadry et al. find out and detail the disparity in stroke awareness
across demographics [9]. Howard et.al. forward the importance of self-reported
or questionnaire-based approach within categorizing the general levels of risk
among the respondents [17]. Luo et al. also demonstrate the impact of stroke
across demographics as reported in the BRFSS using regression model which
checks if there is a relation between the two [10]. Nuyujukian et al. employ logis-
tic regression model to show the association between length of sleeping hours
and stroke across ethnicities [25].
To the best of our knowledge, neither are there any existing work that per-
forms stroke prediction on the basis of the whole BRFSS dataset, nor are there
any such work that focuses on the purely behavioural features. Most of the
4
https://www.cdc.gov/stroke/behavior.htm.
5
https://en.wikipedia.org/wiki/Boosting (machine learning).
Prediction of Stroke Risk Factors for Better Pre-emptive Healthcare 15
other works are either related to finding correlations between certain features
and stroke [19] or they are restricted to only certain states [20]. The downside of
finding only the correlations is that it does not necessarily enable us to hypothe-
size how true we are about the relationship as correlation only denotes a number
to indicate the relationship strength between two variables whether the predic-
tion shows basing ourselves on the predictors how saliently can we hypothesize
that the target is conceivable.
The Class Imbalance Problem and Its Proposed Solution: Apart from
the model selection issues, the class imbalance is a common problem that arises
specially in medical datasets. Moreover, in case of public survey, this is obvious
due to missing values and wrong responses of stroke that is not present in case of
clinical data. Class imbalance often leads the created prediction model to learn
with a bias towards the majority class. For example, if in a dataset the ratio
of label counts between majority and minority class is 25:1 then an accuracy-
driven classifier model may yield an accuracy of more than 90% by disregarding
the impact of minority class instances and classifying all instances as belonging
to the majority class. This problem has been mentioned and worked upon by
Wang and Yao [15], who propose sub-sampling as a way to get better results.
The dataset for this work is taken from the Behavioral Risk Factor Surveillance
System (BRFSS)6 conducted by CDC. Starting in 1984, BRFSS collects data in
all states of U.S. and also the U.S. territories. This health survey is conducted
over telephone and it covers most of the potential health risk factors, health-
related behavioural practices and health conditions. The resulting dataset is
shared with public and is available for free [12]. CDC’s official website7 publishes
the relevant questionnaire and detailed dataset encoding for a very detailed and
comprehensive understanding. In the present work, we use BRFSS 2012 dataset
which records 475687 observations (See Footnote 7).
4 Feature Selection
6
https://www.cdc.gov/brfss/index.html.
7
https://www.cdc.gov/brfss/annual data/annual 2012.html.
Another random document with
no related content on Scribd:
Ludwig und Bodak — lauter tüchtige, prächtige Männer, die für solch
eine Reise wie geschaffen waren.
Der Himmel sah bedenklich aus, nur ein einziger Stern blitzte
zwischen den zerfetzten Wolken. Ich lag noch eine Weile wach und
lauschte der wunderbaren Stille der Wüste, die nur hier und da von
Hundegebell in der Ferne oder von den Lauten eines Nachtvogels
unterbrochen wurde. Ich hörte meine Uhr unter dem Kopfkissen
ticken und die langen, tiefen Atemzüge meiner beiden
Reisekameraden. Dann schlief auch ich ein.
Gegen 11 Uhr aber erwachte ich durch einen furchtbaren
Sturzregen. Schwere Tropfen platschten draußen in neu
entstandene Wasserlachen und trommelten auf die Zeltbahnen. Und
das Trommeln dauerte die ganze Nacht mit unverminderter
Heftigkeit! So oft ich erwachte, hörte ich dieses trostlose Rauschen,
das uns mindestens auch für den nächsten Tag den Weg nach
Osten versperren mußte.
Am Morgen spannten wir bei immerfort strömendem Regen die
Zelttür zu einem Dach auf, das von zwei Stangen getragen wurde,
und frühstückten auf dieser luftigen Veranda Eier und Schinken, die
wir auf einem Primuskocher gebacken hatten. Nach Norden zu lag
die freie Steppe vor uns, aber von den nahen Bergen war keine Spur
mehr zu sehen. Schwere Wolken schwebten über der Erde, und wer
das Zelt verließ, sank fußtief in den roten Schlamm ein.
Gegen Mittag hörte der Regen eine Weile auf. Da kamen auch
schon die Dorfleute neugierig heran, und ein altes runzliges Weib
bat um Medizin für ihren Sohn, den seit einigen Tagen starker
Kopfschmerz und Fieber plagten. Dr. Reith gab ihr etwas für den
Kranken. Vor kurzem war in Bir-dava der Flecktyphus ausgebrochen
und hatte von den fünfunddreißig Einwohnern sieben weggerafft. Für
Geld und gute Worte brachte man uns etwas armseliges Reisig zu
einem Feuer, dessen Glut wir in einem eisernen Kessel ins Zelt
trugen, um in der feuchten Kälte nur ein gewisses Gefühl von
Wärme und Trockenheit zu gewinnen. Kleine zerlumpte Kinder
sammelten Konservenbüchsen und leere Flaschen, und die
Dorfhunde wurden immer frecher auf der Suche nach Abfällen.
Sogar eine Maus quartierte sich unter unsern Kisten und Säcken ein
und entwischte uns immer wieder, so oft wir auch Jagd auf sie
machten.
Schließlich begann es wieder zu regnen. Man hörte und sah, wie
ein Regenschauer nach dem andern wolkenbruchartig über die
Steppe daherkam. Sogar in unserm Zelt waren wir nicht mehr sicher.
Die Chauffeure mußten ringsherum einen Kanal graben und einen
Erdwall aufwerfen, um uns vor Überschwemmung zu schützen. Vor
dem Eingang bauten sie eine Brücke aus Planken des
Lastautomobils. Das ganze Feld war ein einziger Sumpf, denn es
dauerte lange, bis das Regenwasser den Lehm durchdrang, dessen
Oberfläche glatt war wie Seife. Ein tragikomisches Schauspiel bot
eine vorüberziehende kleine Karawane von Mauleseln: die auf den
schwer beladenen Saumsätteln sitzenden Araber hatten ihre
schwarz- und braungeränderten Mäntel über den Kopf gezogen, und
der Regen floß nur so von ihnen und ihren Tieren herunter. Die
endlich hereinbrechende Dämmerung wirkte fast wie eine Erlösung.
Wir bereiteten mit möglichster Langsamkeit unser Abendessen und
überließen uns einer neuen Nacht.
Gegen Mitternacht weckte mich wiederum ein fast tropischer
Regenguß aus dem Schlaf. Feine Wassertropfen sprühten durch das
Zelttuch auf uns herab, und einige Stunden später regnete es so
kräftig herein, daß Dr. Reith in das Unwetter hinaus mußte, um den
Schutzlappen des Zeltfensters an der Windseite wieder
festzubinden, der aufgegangen war.
Am Tage wurde es nicht besser. Die Zelttür mußte geschlossen
bleiben, denn der Wind stand gerade gegen sie; wir frühstückten auf
meinem Bett und saßen da wie Schiffbrüchige auf kleiner Klippe in
einem Meer von Schlamm; der Zeltgraben stand ebenfalls bis an die
Ränder voll Wasser. Die schmutzigen Hunde wurden immer kecker,
da wir sie nicht verfolgen konnten. Dazu kam die erschreckende
Nachricht, daß der Kranke von gestern über Nacht gestorben sei —
wir hatten also den Flecktyphus als nächsten Nachbarn!
Später am Tage sahen wir denn auch, wie sich ein kleiner
Leichenzug nach dem Friedhof bewegte, der in einiger Entfernung
südöstlich von unserm Zeltplatz lag. Auf einer Bahre trug man den
Toten langsam dahin. Am Grabe sprachen die Begleiter
Totengebete; bald reckten sie die Hände in der Richtung nach
Mekka empor, bald sanken sie in dem fürchterlichen Schlamm
neben der Leiche nieder. Endlich wurde der Tote in die Erde
gesenkt, abermals Gebete gesprochen und das Grab zugeschaufelt.
Ich glaube, die Zeremonie dauerte ein paar Stunden. Dabei regnete
es unaufhörlich, und die groben Mäntel der Fellachen glänzten von
Wasser. Nach vollbrachter Arbeit ging das Trauergefolge ebenso
langsam nach Haus, wie es gekommen war.
Während der Flecktyphus-Epidemie in Aleppo haben deutsche
Ärzte die Beobachtung gemacht, daß Europäer für Ansteckung weit
empfänglicher sind als Eingeborene. Auch hat sich während des
Krieges gezeigt, daß die Krankheit bei russischen Soldaten und
Kosaken einen milden Verlauf nimmt, weil sie an Ungeziefer
gewöhnt sind. Je älter der Patient ist, um so schwerer kommt er
durch.
Einem türkischen Soldaten, der gen Westen ritt, gab der Major
einen Brief mit, worin er die zurückgelassenen Chauffeure über
unsre Lage unterrichtete. Sonst zeigte sich kein Reisender, der kühn
genug gewesen wäre, den Kampf mit den Elementen aufzunehmen.
Bei Dunkelwerden hörte der Regen auf. Da brachte „Lohengrin“
— wie Lundgren von den Kameraden genannt wurde — das
Teewasser ins Zelt und meldete, im Norden sei eine Reihe Feuer
sichtbar. Was mochte das sein? Biwakfeuer? Doch nicht etwa
russische? Aber das war unmöglich, dann hätten wir etwas von
einem Rückzug der Türken merken müssen. Diese waren aber im
Vormarsch. Nach der Karte lag in jener Richtung, nur 10 Kilometer
entfernt, Mardin, das sich seit gestern hartnäckig hinter dem
Regenschleier verborgen hatte; die Feuer waren nichts anderes als
die Lampen in den Häusern dieser Stadt. Wir konnten uns also ohne
Sorge in unserm Gefängnis zur Ruhe begeben und den
Regenschauern lauschen, die am Abend mit vermehrter Heftigkeit
einsetzten.
Am nächsten Tag dasselbe Bild! Man steht auf, wäscht sich,
kleidet sich an, öffnet einen Spalt der Zelttür, frühstückt und hat den
ewig langen Tag vor sich. Ich habe Bezolds „Ninive und Babylon“
und „Moltkes Briefe aus der Türkei 1835–39“ bei mir, aber die Ruhe
zum Lesen fehlt. Man wartet in sehnsüchtiger Qual, daß irgendetwas
geschehe, uns aus dieser hoffnungslosen Lage zu befreien.
Volkstypen zu zeichnen ist auch unmöglich; Leute mit Flecktyphus
bringenden Läusen ins Zelt hereinlassen — das fehlte noch! Im
Dorfe geht das Leben seinen alltäglichen Gang. Frauen treiben von
den Feldern Schafe und Ziegen herein oder holen in Lehmkrügen
Wasser am Brunnen. Der Himmel ist blauschwarz. Zuweilen grollt
unglückverheißend und dumpf der Donner in den Bergen. Und das
ist Mesopotamien im April, wo ich Frühlingswärme erwartet hatte,
Trockenheit und Skorpione! Aber wir waren ja freilich in einer Höhe
von 550 Meter und am Fuß eines Gebirges, wo der Winter noch
nicht gewichen war. Nachmittags um 5 Uhr zeigte das Thermometer
nur 10 Grad, eine Temperatur, die wir nach den warmen
Sonnentagen in Aleppo als Kälte empfanden.
Drei Soldaten kamen von Ras-el-Ain zu Fuß; sie hatten unsre
verunglückten Automobile und das Zelt der Chauffeure gesehen. Ein
paar andere zogen in entgegengesetzter Richtung. Sie waren schon
20 bis 30 Kilometer östlich von Nesibin in das Moorbad geraten, das
bis hierher reichte, und sie gaben uns die tröstliche Versicherung,
daß mindestens zwei Tage warmer Sonnenschein nötig seien, um
das Land wieder zu trocknen.
Am Nachmittag trat einen Augenblick die Sonne hervor, und mit
ihr in Nordnordwest die alte Festung von Mardin auf dem Gipfel des
Bergkammes; unmittelbar darunter die Häuser wie Schwalbennester
an den Böschungen, dazwischen die armenischen und syrischen
Kirchen und weißen Minarette. Aber bald verschwand wieder alles
unter schwarzen Wolkenmassen und neuen Regenschauern.
Eine verzweifelte Lage! Wären wir nur einen Tag früher
aufgebrochen, so wären wir bereits in Mosul! Die ganze Strecke ist
nur 320 Kilometer lang, für ein Auto zwei Tage Fahrt. Nun saßen wir
in diesem elenden Gefängnis und konnten weder vor- noch
rückwärts. Proviant hatten wir ja noch für acht Tage, nur Brot und
Wasser gingen zu Ende. Aber an letzterem war ja kein Mangel — wir
brauchten nur ein Stück Segeltuch aufzuspannen, um die Kannen
gefüllt zu erhalten.
Am Morgen des 3. April weckte uns die Meldung, die Sonne
scheine. Wirklich! Der halbe Himmel blau und hell, und über die
andere Hälfte segelten freundliche weiße Frühlingswolken. Wir
kleideten uns in aller Eile an und rasierten uns sogar aus lauter
Feststimmung. Für vier Soldaten, die von Ras-el-Ain
dahergewandert kamen, kauften wir bei der Dorfbevölkerung einige
Brote, denn sie hatten nichts mehr zu essen, da sie während der
Regentage hatten liegen bleiben müssen.
Viertes Kapitel.
Mein neuer Feldzugsplan.
Fünftes Kapitel.
Auf den Wellen des Euphrat.