Final Seminar Rep

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 25

VISVESVARAYA TECHNOLOGICAL UNIVERSITY

“JNANA SANGAMA”, BELAGAVI - 590 018

A SEMINAR REPORT
on

“Predicting Anxiety and Depression In Elderly


Patients Using Machine Learning Technology”

Submitted by

Keerthana Rajeev 4SF16CS182

In partial fulfillment of the requirements for the VIII semester of

BACHELOR OF ENGINEERING

in

COMPUTER SCIENCE & ENGINEERING

at

SAHYADRI
College of Engineering & Management
Adyar, Mangaluru - 575 007
2019 - 20
SAHYADRI
College of Engineering & Management
Adyar, Mangaluru - 575 007

Department of Computer Science & Engineering

CERTIFICATE

This is to certify that the seminar entitled “Predicting Anxiety and Depression In
Elderly Patients Using Machine Learning Technology” has been successfully pre-
sented out by Keerthana Rajeev (4SF16CS182), the bonafide students of Sahyadri
College of Engineering & Management in partial fulfillment for the VIII semester of Bach-
elor of Engineering in Computer Science & Engineering of Visvesvaraya Technological Uni-
versity, Belagavi during the year 2019 - 20. It is certified that all corrections/suggestions
indicated for Internal Assessment have been incorporated in the report deposited in the
departmental library. The seminar report has been approved as it satisfies the academic
requirements as per university guidelines.

———————————– ——————————— ————————————–


Seminar Coordinator 1 Seminar Coordinator 2 Head of the Department
Dr. Pushpalatha K Mrs. Soumya Patil Dr. J V Gorabal

Marks Awarded
Abstract

Anxiety and Depression have proven to be two prominent mental-health problems among
the older demographics. Multitude of morbidities have been rising because these issues are
often undiagnosed. Timely diagnosis at the initial stages can have a profound impact on
the efficient management of anxiety and depression along with the associated disorders.The
capacious load of patients and the time constraints play a huge role particularly in a devel-
oping country like India making it extremely hard to detect both anxiety and depression
based on psychometric tests among the elderly population. Therefore, it is truly important
to produce a predicting model that can aid surgeons in identifying anxiety and depression
effectively. The objective of this paper is to curate a predictive model that is capable of
diagnosing anxiety and depression among the geriatric community subjects on the basis
of factors such as personal and socio-demographic attributes with the essence of machine
learning. Ten different classifiers were used which used a dataset consisting of 510 geriatric
patients and then tested with the ten-fold cross-validation technique. The random forest
classifier was able to obtain the highest accuracy with 89 percent. The external validation
is carried out by testing the RF model obtained with a dataset comprising of110 geriatric
patients. In comparison to the gold standard tool, the predictive accuracy obtained with
the RF model was a 91 percent with a false positive rate of 10 percent.

i
Acknowledgement

It is with great satisfaction and euphoria that I am submitting the Seminar Report on
“Predicting Anxiety and Depression In Elderly Patients Using Machine Learn-
ing Technology”. I have completed it as a part of the curriculum of Visvesvaraya Techno-
logical University, Belagavi for the award of Bachelor of Engineering in Computer Science
& Engineering.

I am profoundly indebted to our seminar coordinators, Dr. Pushpalatha K, Associate


Professor , and Mrs. Soumya Patil, Assistant Professor, Department of Computer
Science & Engineering for innumerable acts of timely advice, constant encouragement,
support and I sincerely express our gratitude.

I express my sincere gratitude to Dr. J V Gorabal, Head & Professor, Department of


Computer Science & Engineering for his invaluable support and guidance.

I sincerely thank Dr. R Srinivasa Rao Kunte, Principal, Sahyadri College of Engi-
neering & Management, and Dr. D. L. Prabhakara, Director, Sahyadri Educational
Institutions,who have always been a great source of inspiration.

Finally, yet importantly, I express my heartfelt thanks to my family & friends for their
wishes and encouragement throughout the work.

Keerthana Rajeev (4SF16CS182)

ii
Table of Contents

Abstract i

Acknowledgement ii

Table of Contents iv

List of Figures v

List of Tables vi

1 Introduction 1

2 Literature Survey 2

3 Problem Definition 4

4 System Design 5
4.1 Flow diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

5 Implementation 6
5.1 DATA COLLECTION PHASE . . . . . . . . . . . . . . . . . . . . . . . . 6
5.1.1 Data collection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
5.1.2 Hospital ADS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
5.2 FEATURE EXTRACTION . . . . . . . . . . . . . . . . . . . . . . . . . . 7
5.2.1 Feature Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
5.2.2 Attribute Evaluators . . . . . . . . . . . . . . . . . . . . . . . . . . 7
5.3 CLASSIFICATION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
5.3.1 Classification in WEKA . . . . . . . . . . . . . . . . . . . . . . . . 8
5.3.2 Classifiers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
5.4 EXTERNAL VALIDATION . . . . . . . . . . . . . . . . . . . . . . . . . . 10

iii
5.4.1 External validation and Gold Standard comparison . . . . . . . . . 10

6 Results and Discussion 11


6.1 Attribute Selection and Classification . . . . . . . . . . . . . . . . . . . . . 11
6.1.1 Feature Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
6.1.2 Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
6.2 External Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
6.2.1 Gold Standard . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
6.2.2 Evaluation of RF classifier . . . . . . . . . . . . . . . . . . . . . . . 15

7 Conclusion and Future work 16

iv
List of Figures

4.1 Flow diagram of Methodology . . . . . . . . . . . . . . . . . . . . . . . . . 5

5.1 Methodology in brief with features, classifiers and attribute evaluators . . . 9

6.1 The confusion matrix obtained after cross-validation . . . . . . . . . . . . . 13


6.2 A bar graph diagram depicting the four different metrices . . . . . . . . . . 14
6.3 Displays the confusion matrix for the HADS standard test . . . . . . . . . 15

v
List of Tables

6.1 Comparison of five different attribute evaluator algorithms . . . . . . . . . 11


6.2 Evaluation of classifiers with performance metrics . . . . . . . . . . . . . . 13
6.3 The metrics after evaluation by RF classifier . . . . . . . . . . . . . . . . . 15

vi
Chapter 1

Introduction

The demographics across the world are experiencing an enormous rise over the past few
decades. The population was said to be a 607 million in the year 2000 and stood at
901million in the year 2015. The World Health Organisation clearly states that a healthy
mind combined with a healthy body culminates in a healthy body. When alterations occur
respect to these aspects, problems such as anxiety and depression arises. The geriatric
population suffers from both these morbidities more than the average age group which
leads to further fatalities.
In India, the existence of anxiety and depression are worthy of mentioning and con-
tributes majorly with a 12percent of the entire global burden and is projected to increase
in the following years. Although this can be prevented, the diagnosis in the initial stages
prove to be extremely difficult and if failed to do so can create panic and worsen the
disorder to a huge extent thereby contributing to further casualties.
The Hospital Anxiety and Depression Scale is a well known psychometric test that
prompts the user to answer 14 question which ultimately decides whether he is suffering
from anxiety and depression or not. But in a country like India where the population is
immense, it is extremely hard to conduct HADS test by a surgeon or physician.
Therefore, a predictive model which uses the best classifier was developed with the
purpose of making the entire process much more efficient thereby increasing the accuracy
in comparison to the previous test and the problems regarding various time constraints
can be prevented. This model can be used to easily classify the person as having anxiety
and depression or not.

1
Chapter 2

Literature Survey

In [1] Lopez et al. presented a system for diagnosing the most commonly occurring disease
Alzheimer’s, an automated system was used to interpret the positron emission tomogra-
phy and single photon emission computed tomography pictures. The biggest issue to be
addressed here is the size of the sample which is extremely small comprising of less number
of existing images when comparison to the higher number of features. This drawback can
be overcome by extensively limiting the dimension of the feature space with the help of
principal component analysis (PCA). The approach mentioned here makes use of Bayesian
Classifier which utilized posteriori knowledge to understand which class the data point
belongs to with an accuracy 88.6 and 98.3 percent for both SPECT and PET images.
There was a huge improvement in the accuracy values with respect to the already existing
methods.
In [2] Ansari et al. proposed a system to diagnose and classify patients who are suffering
from hepatotropic viral disease using a neural network based approach. This technique
uses the real time data which is in the form of liver tests and virological markers.The neural
network which was trained processes the unknown patients which were not present in the
training dataset. The approach used here has clearly proved that the results it yielded
was able to diagnose the disease as well as classify the former to various different stages
namely chronic or acute.The performance of the neural network is highly accurate even for
a small dataset where as the Kohonen’s unsupervised maps does not do effectively for the
task having similar criterias.
In [3] Lashkari et al. presented a paper that deals with a automated technique that
assists the surgeons to detect breast cancer at an early stage depending on the vari-

2
Predicting Anxiety and Depression In Elderly Patients Using Machine Learning Technology Chapter 2

ous degrees. Initially,the interested region is found out with the help of automated op-
erations and the image quality is further enhanced.Later, various features such as his-
togram,statistical,frequency domain and morphological aspects are extracted from the left
as well as right segmented breasts.To acquire the best features and enhance the accuracy
of this method, feature selection methods are used along with genetic algorithms. Ulti-
mately, supervised methods namely SVM,AdaBoost, Knearest neighbour and Naı̈ve Bayes
are applied to classify as well as TH labeling steps.The results was able to produce a good
mean accuracy of about 88.03 percent for using 0.
In [4] Study of data mining and analytics is the quest for useful knowledge in large data
volumes. It is now commonly used in the health-care field. Breast cancer in particular is the
second most dangerous creator of cancer, and the second most dangerous. The only way to
increase the chances of long-term survival for a survivor of breast cancer is to detect it as
early as possible.There are currently three types of breast cancer diagnosis: mammography,
FNA (fine needle aspirate), and surgical biopsy. Mammography’s diagnostic accuracy
varies from 68 to 79 percent, FNA’s accuracy is inconsistent with ranging from 65 to 98
the accuracy of an active biopsy is almost 100.Nevertheless, the surgical biopsy technique is
both disagreeable and expensive. In this paper we use a FNA with a system of data mining
statistics to get an simple way to get the best result. To find the unsuspected relationships,
we combine several statistical approaches such as PCA, PLS linear regression analysis with
data mining approaches such as select attribute, decision trees, and association laws.
In [5] Bhakta et al.came up with a ANN model to identify depression. People above the
age of 60 years or more are classified into the category of geriatric population. And among
this section of people, depression appears to be a widely occurring disability. The aspect
of brain aging cannot be acknowledged as physiological and an invinsible part of the whole
aging process even though it is highly vulnerable to anxiety and depression.Therefore,
depression can be diagnosed by using ANN model from various different factors.

Department of Computer Science and Engineering, SCEM, Mangaluru Page 3


Chapter 3

Problem Definition

To build an automated disease prediction system that should aid the general physicians
and specialists in all fields of specialization for screening of the patient for both anxiety
and depression effectively as well as the underlying diseases.

4
Chapter 4

System Design

4.1 Flow diagram

Figure 4.1: Flow diagram of Methodology

5
Chapter 5

Implementation

5.1 DATA COLLECTION PHASE

5.1.1 Data collection

R.G. Kar Medical College and Hospital located at West Bengal is the source of the dataset
obtained and used in this paper. The dataset of 520 geriatric patients were taken from
the hospital for a duration starting from January to August 2016. Information consisting
of factors such as sex,marital status,age,residence,bereavement,personal income,history of
depression and anxiety,insomnia,mobility impairment ,literacy, residence, marital status
were taken from 520 patients for examination. These attributes were chosen after con-
sulting with leading psychiatrists and various literature surveys. The former represented
various causative dimensions of depression and anxiety among the elderly. The various at-
tributes collected belongs to 5 different categories which can further sub categories. These
include economic conditions,personal attributes,current medical conditions, past medical
history and socio-demographic factors which in turn comprises of the above mentioned fac-
tors including sex and age, financial conditions, past conditions of anxiety or depression.
Once these factors were determined, they were evaluated using the hospital anxiety and
depression scale (HADS).

5.1.2 Hospital ADS

HADS is a psychometric analysis test for determining anxiety and depression in the patients
recommended by doctors. The test comprises of 14 queries out of 7 was used to evaluate

6
Predicting Anxiety and Depression In Elderly Patients Using Machine Learning Technology Chapter 5

depression and the remaining seven to find out anxiety. In an ordinal scale, responses were
marked from a mark ranging from 0 to 3. A patient can clearly score anything from a range
within 0 to 27 for anxiety and depression. Various different articles and sources stated that
the cut-off point was an 8 out of the overall score of 21 for rating anxiety or depression. This
scale proved to act as a rudimentary standard for patients suffering from the disorder and
it takes from about 5 to 30 minutes to complete and finish this questionnaire. Based on the
score secured, the four outputs were namely Depression/Anxiety present or absent,Only
depression and Only anxiety. Finally, these four results were further narrowed down into
two categories that is Depression and Anxiety present or absent and based on these results,
the doctors were able to classify the patient easily.

5.2 FEATURE EXTRACTION

5.2.1 Feature Selection

The feature selection process is carried out to reduce the feature dimension and redundant
data. In the beginning, 20 features were narrowed down to predict the output but its
essential to reduce the number of features and include those which can efficiently classify
the patients to the two categories. Since the issue of interaction among the various features
and multicollinearity occur, it can have an adverse effect on the accuracy of the learning
algorithm. So for reducing the dimensionality and redundancy, various(5) attribute evalu-
ators were used with a filter mechanism. Ultimately, the group of attributes that best fits
were determined and proceeded for further predictive modelling.

5.2.2 Attribute Evaluators

For evaluating the various factors, 5 different evaluators were used. Correlation feature
selection evaluator carries out the evaluation process by taking into account the predictive
capacity of a single attribute with the redundancy that exists between them. Collinearity
being less is a preference along with set of features which are greatly linked to the class.
One R (OR) performs evaluation of a feature by making use of an OR classifier which
produces easy rules for each individual attribute and then chooses the rule having the
least error. Principal component analysis evaluates an attribute by selecting required

Department of Computer Science and Engineering, SCEM, Mangaluru Page 7


Predicting Anxiety and Depression In Elderly Patients Using Machine Learning Technology Chapter 5

number of eigen vectors to contribute to a little percent of variance present in the actual
data thereby reducing the features dimensions. Gain ratio and symmetrical uncertainty
evaluators carries out the evaluation steps on the basis of entropy measurement of a system
which can measure the uncertainty of the system.

5.3 CLASSIFICATION

5.3.1 Classification in WEKA

Weka is a combination of multiple algorithms aimed to solve real-world problems.This is


a binary classification problem where the output is either Depression/Anxiety present or
Depression/Anxiety absent. Various machine learning algorithms can be used for health-
related classification problems effectively. In this approach,ten varieties of classifiers were
employed and evaluation was performed to choose the most optimal one among them.
The classifiers are Logistic, Bayesian Network, Naı̈ve Bayes, Random Forest,J48, Multi-
ple Layer Perceptron, Sequential Minimal Optimization, K-Star,Random Tree ,Random
Sub-Space. These various classifiers are based on Regression technique,Artificial Neural
Network, Bayesian technique, instance based method and many more.

5.3.2 Classifiers

Ten different classifiers were used to evaluate the predictive variables to predict the output.
Bayesian Network classifier uses a DAG to represent the data and relationship existing
between the various node uses the conditional probability techniques. Finally the network
can be trained on the basis of the relationship existing between the entire network and
the nodes. Logistic classification can be used when the output variables are categorical
in nature that is each and every single attribute makes a contribution for predicting the
essential outcomes from the dataset. Maximum-likelihood estimation is used for calculating
the predictive probability of the variable. Finally a binary output is produced based on the
probability of prediction. Multiple Layer Perceptron is a simple and very basic artificial
neural network that is used for supervised learning. This system consists of numerous
number of inputs and an individual output. This output relies on the weighted sum of the
various input node. In complex situations, input of one layer will have to be fed to the next

Department of Computer Science and Engineering, SCEM, Mangaluru Page 8


Predicting Anxiety and Depression In Elderly Patients Using Machine Learning Technology Chapter 5

day and so on. The output layer will use an activation function which will be activated when
the input value exceeds a certain threshold. The hidden layer and output layer node number
depends solely on the application. Naive Bayes classifier uses the probability statistics and
is an easy supervised learning method wherein the probability is computed using the Naı̈ve
Bayes algorithm. Random Tree classifier builds either a DAG or a tree on the basis of a
stochastic process that selects K randomly selected variables at each of the nodes. It gives
us an option to provide estimating the class probabilities. Random Forest is an assembly
of random decision tree classifier. The predictive output of the RF.classifier is generated
by the outputs predicted by all the individual classifiers. The base learner employed
in the Random Forest is Random Tree. The Random Sub-Space method constructs a
decision tree-based classifier by using reduced error pruning. It is a meta classifier used.
Final classification result is on the basis of majority voting principal. Sequential Minimal
Optimization is an efficient method used to solve quadratic programming problems easily.
It divides the large problem into smaller ones and solves them separately with less memory
requirements. J48 classifier performs a tree-based classification by making using of C4.5
decision algorithm. On the basis of the group of values for the set of attributes, the input
data is split into different trees. K-Star is one of the instance-based classifiers. The test
instance’s class is determined on the basis of training instances that are identical with
its class. What makes it strikingly different from other instance based methods is that it
makes use of an entropy based distance formula to calculate the similarity in the figure
5.1.

Figure 5.1: Methodology in brief with features, classifiers and attribute evaluators

Department of Computer Science and Engineering, SCEM, Mangaluru Page 9


Predicting Anxiety and Depression In Elderly Patients Using Machine Learning Technology Chapter 5

5.4 EXTERNAL VALIDATION

5.4.1 External validation and Gold Standard comparison

A test data collection consisting of knowledge of the previously selected ten features after
the feature selection step is garnered from 110 elderly patients of the same hospital from
a duration ranging from September to October of 2016. Hand in hand, the levels of
depression and anxiety were computed for those patients using the HADS test and the
output was recorded for the upcoming stage. The outcome of this phase was called as a
gold standard.
A data set was curated which comprised of only the ten chosen features which has
missing decision class. At inception, previously made RF classifier model was fed into
WEKA, and the test data was provided and the former model was re-evaluated on the
testing data. Class prediction was carried out for each of these instance and the resultant
output was captured for comparing with the existing gold standard.

Department of Computer Science and Engineering, SCEM, Mangaluru Page 10


Chapter 6

Results and Discussion

6.1 Attribute Selection and Classification

Among the 510 elderly community subjects that was selected for this approach, there were
281 identified male patients and 229 female patients which contributes 55 percent and 45
percent of the entire selected population. The Mean that is (+,-)standard deviation age
was found to be 68.5 years. It was concluded that 48.2 percent of these patients were
suffering from mental health issues which being either depression or anxiety or both the
former and latter and 51.8 percent of the patients were found to be freed of these mental
disorders.

Table 6.1: Comparison of five different attribute evaluator algorithms

ORDER CFS best Fit One R PCA Gain Ratio Symmetrical Uncertainty
1 A A A A A
2 S S P S S
3 M E R E P
4 E R E R R
5 R M S S E
6 P P D HTN S
7 D H M D D
8 HTN HTN V V PI
9 H D H H HTN
10 V V HTN PI V

11
Predicting Anxiety and Depression In Elderly Patients Using Machine Learning Technology Chapter 6

6.1.1 Feature Evaluation

Various different attribute evaluators were applied to the features selected to reduce the
dimensions of the features. After applying the 5 different algorithms, the results obtained
are displayed in the table below. Careful examination and contemplation of the different
evaluators, left us out with 11 features out of the overall 20 features which were initially
chosen. These factors were namely sex, personal income, age, marital status, HTN, pain
at multiple sites, recent bereavement, hearing impairment, employment status, diabetes
mellitus, and visual impairment. Features were then passed on to the classifiers.

In an developing country like India, PI or person income ceases to be a piece of infor-


mation that is often unreliable due to various factors and people often hesitate to mention
the exact figure due to various socio,cultural and economic aspects.Therefore, this feature
has been eliminated in the further classification process.

6.1.2 Classification

The ten selected features are then provided into the classifiers which are of 10 types. The
figure in clearly depicts the classifications carried out with the 520 geriatric patients that is
the confusion matrix generated after the 10-fold cross validation method. The variables a
and b stands for anxiety and/or depression present and anxiety and depression absent. The
kappa value is the measure of how well the variables classified by the classifiers matches
with that those of data marked as the ground truth.

Furthermore, the classifiers are evaluated in terms of various performance metrics such
as true positive rate (the proportion of positives which are labelled as such), accuracy(The
ratio of right predictions to the overall number of input samples), false positive rate(The
outcome where the model predicts the positive class to be incorrect), positive predictive
value (Quantity of actual positives classified as positives), area under the receiver operat-
ing characteristic curve (ROC represents a probability curve where AUC is the degree of
separability). Higher the value of this term, the better is the model at making accurate
predictions and f-measure( It is the harmonic mean of the values obtained from precision
and recall). The figure in 6.2 shows the classifiers with the performance factors.

Department of Computer Science and Engineering, SCEM, Mangaluru Page 12


Predicting Anxiety and Depression In Elderly Patients Using Machine Learning Technology Chapter 6

Table 6.2: Evaluation of classifiers with performance metrics

Classifier Accuracy TP FP PPV F-Measure Area Under Curve


BN 79.8 79.8 15.5 81.8 79.7 88.9
NB 79.6 79.6 19.6 81.4 79.4 85.3
LOG 72.4 72.4 27 73.4 72.2 81.1
MLP 77.8 77.8 22.1 77.9 77.8 85
SMO 75.3 75.3 23.4 79.7 74.6 75.9
KS 75.3 75.3 24.7 75.3 75.3 81.4
RS 87.5 87.5 12.4 87.5 87.5 91.7
J48 87.8 87.8 12 87.9 87.8 86
RF 89 89 10.9 89.1 89 94.3
RT 85.1 85.1 15 85.1 85.1 85

Figure 6.1: The confusion matrix obtained after cross-validation

Department of Computer Science and Engineering, SCEM, Mangaluru Page 13


Predicting Anxiety and Depression In Elderly Patients Using Machine Learning Technology Chapter 6

The above table clearly indicates that Random Forest is the most appropriate classifier
for the selected data set with an accuracy of about 89 percent (highest among all the other
classifiers), AUC (94.3). F-measure (89), TP rate (89) and PPV (89.1). The false positive
rate is the lowest at 10.9 percent. A bar graph representation is depicted 6.2 showing the
same statistics.

Figure 6.2: A bar graph diagram depicting the four different metrices

6.2 External Validation

6.2.1 Gold Standard

A test data collection consisting of knowledge of the previously selected ten features after
the feature selection step is garnered from 110 elderly patients of the same hospital from
a duration ranging from September to October of 2016. Hand in hand, the levels of
depression and anxiety were computed for those patients using the HADS test and the
output was recorded for the upcoming stage. The outcome of this phase was called as a
gold standard. The below table in the figure 6.3 shows the confusion matrix for prediction
by the RF classifier in comparison to the HADS standard on the testing data.

Department of Computer Science and Engineering, SCEM, Mangaluru Page 14


Predicting Anxiety and Depression In Elderly Patients Using Machine Learning Technology Chapter 6

Figure 6.3: Displays the confusion matrix for the HADS standard test

6.2.2 Evaluation of RF classifier

A data set was curated which comprised of only the ten chosen features which has missing
decision class. At inception, previously made RF classifier model was fed into WEKA,
and the test data was provided and the former model was re-evaluated on the testing
data. Class prediction was carried out for each of these instance and the resultant output
was captured for comparing with the existing gold standard. The diagram in the figure 6.3
represents the accuracy obtained by the RF classifier.The prediction accuracy had increased
and was calculated to be a 91 percent with a false rate of 10 percent in comparison to gold
standard.

Table 6.3: The metrics after evaluation by RF classifier

Metric Number Metrics Performance

1 Accuracy 91
2 TP Rate 92
3 FP Rate 10
4 Precision 89
5 Specificity 90
6 F-Score 90
7 Diagnostic odds ratio 104

Department of Computer Science and Engineering, SCEM, Mangaluru Page 15


Chapter 7

Conclusion and Future work

This approach clearly depicts that machine learning technologies can be widely deployed
in medical fields for automated timely prediction of illness. By making use of the existing
techniques, by carrying out a comparison among the various classifiers we can easily build
a prediction model for research and medical centers comprising of multitudes of data from
numerous sources. This will enable us to build a more generalized model that is efficient
and effective for classification of depression and anxiety as accurately mentioned in this
technique.
Furthermore, this can aid the physicians and doctors to accommodate and focus on
more health-issues emerging in the upcoming future with proper emphasis and under-
standing especially and particularly in the case of the geriatric population.

16
References

[1] M. Lopez, J. Ramirez, J. M. Gorriz, D. Salas-Gonzalez, I. A. Lvarez, F. Segovia,


and C. G. Puntonet, “Automatic tool for alzheimer’s disease diagnosis using pca and
bayesian classification rules,” Electronics Letters, vol. 45, pp. 389–391, April 2009.

[2] S. Ansari, I. Shafi, A. Jamil, and S. Shah, “Neural network-based approach for the
non-invasive diagnosis and classification of hepatotropic viral disease,” IET Communi-
cations, vol. 6, pp. 3265–3273, 12 2012.

[3] A. Lashkari, F. Pak, and M. Firouzmand, “Breast thermal images classification using
optimal feature selectors and classifiers,” The Journal of Engineering, vol. 2016, no. 7,
pp. 237–248, 2016.

[4] Xiangchun Xiong, Yangon Kim, Yuncheol Baek, Dae Wong Rhee, and Soo-Hong
Kim, “Analysis of breast cancer using data mining statistical techniques,” in Sixth
International Conference on Software Engineering, Artificial Intelligence, Networking
and Parallel/Distributed Computing and First ACIS International Workshop on Self-
Assembling Wireless Network, pp. 82–87, May 2005.

[5] I. Bhakta and A. Sau, “Prediction of depression among senior citizens using machine
learning classifiers,” International Journal of Computer Applications, vol. 144, pp. 11–
16, 06 2016.

You might also like