Proceedings of the Third International Conference on Inventive Research in Computing Applications (ICIRCA-2021)

IEEE Xplore Part Number: CFP21N67-ART; ISBN: 978-0-7381-4627-0

Machine Learning Techniques for Prediction of Mental Health

1
Tarun Jain,2 Ashish Jain,3 Priyank Singh Hada, 4 Horesh Kumar, 5 Vivek K Verma, 6 Aayush patni
1
[email protected] m,2 [email protected] m,3 [email protected],4 horeshkiet1991@g mail.co m,
5 6
Vermavivek123@g mail.co m, [email protected],
[1,2,3,5,6]
SCIT, Manipal University Jaipur, 4 GNIOT Group of Institutions Greater Noida, Uttar Pradesh, India
2021 Third International Conference on Inventive Research in Computing Applications (ICIRCA) | 978-1-6654-3877-3/21/$31.00 ©2021 IEEE | DOI: 10.1109/ICIRCA51532.2021.9545061

Abstract— Suicide is the 2 nd leading cause of death in the world, for those aged 15-24 and about 800,000
victims of suicide yearly (all age), which is about 40 per second. Behavioural health disorder, explicitly
depression, are the type of health concerns, not many are aware of. There is no way one can get treatment
of something they are not aware of. So, classifying potential health disordered person is the first step
towards prevention. Lifestyle is something which defines individual the best. Lifestyle including Income,
age group, martial status, child, property owned, alcohol or tobacco consumption, medical expenditure,
insurance or other type of investment and many more. Using 76 such kind of attributes, model will
predict if the individual is victim of depression or not. The proposed model has used eight mainstream
ML calculation methods, namely (Decision tree (DT), Random Forest(RF), Support Vector
Machine(SVM), Naïve Bayes(NB), Logistic Regression(LR), XGBoost(XGB), Gradient Boosting
Classifier(GBC) and Artificial Neural Network(ANN) to build up the expectation models utilizing a huge
dataset (1429 individual’s survey), bringing about precise and productive dynamics. By using various
strategies and different model, this research work has attempted to get a clear and precise picture. The
reason to follow various approaches is that, precise the information, work in a better way and reduce the
number of suicide case. The final outcome received was 87.38 percent, which was using Support Vector
Machine (SVM).

INTRODUCTION Being stress free is one of the key components for

healthy life [9].
Mental health is one of the leading causes to Goal was to make a ML model, using which the
suicide [10]. As most of the victims know a little future possibility of mental health can be identified.
about their mental health, it goes unseen. Every Using and analysing all the data in logical manner
year more than 800,000 individuals die due to gave us insight that there are some similarities in
Depression or some mental health disorder and this all these cases [2]. Using which the mental health
number is not a small one or which could be can be predicted and using the same model, the
ignored [5]. Most of the individuals are of the age potential mental health victim can also be
between 15-24. Which is mostly youth and have a predicted.
lot of potential to become something and give The utilized dataset has 1429 tuples and 76
society a lot [1]. To achieve success in such cases , attributes (Some attributes: name, surveyid, village,
the potential cases must be sorted before even they survey_date, femaleres, age, married, children,
know. Mental health has a lot to do with the hhsize, edu)
amount earned and the way it has been spent [8]. The eight classification algorithms used to classify
Our spending and liabilities tell a lot about our such type of mental conditions are:
potential mental health. examining previous cases,
Decision tree(DT),
in which it has considered how much individual do
Random Forest(RF),
earn and there spending habits or where there most Support Vector Machine(SVM),
chunk of money goes, if they use tobacco, alcohol,
Naïve Bayes(NB),
medical expenses, Social expenses, Education
Logistic Regression(LR),
expenses, number of time they eat meat or fish, XGBoost(XGB),
slept hungry or not, enough food for tomorrow,
Gradient Boosting Classifier(GBC)
Doctor consulting and other kind of liabilities. Artificial Neural Network(ANN)
Also, the family structure has been considered, in
that how many members are there in family, age of METHODOLOGY
the respondent, if the respondent is female, Marital
status, number of children. Assets are also The dataset has been acquired from both the
included, such as household size, number of farms, individuals, who are affected and not affected with
value of live stocks, value of durable goods, value such a condition. Our approach includes utilization
of savings, food own production and many more. of arrangement systems like (Decision tree (DT),
Random Forest (RF), Support Vector Machine

978-0-7381-4627-0/21/$31.00 ©2021 IEEE 1606

Authorized licensed use limited to: JAWAHARLAL NEHRU TECHNOLOGICAL UNIV. Downloaded on March 14,2024 at 10:03:23 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Third International Conference on Inventive Research in Computing Applications (ICIRCA-2021)
IEEE Xplore Part Number: CFP21N67-ART; ISBN: 978-0-7381-4627-0

(SVM), Naïve Bayes (NB), Logistic Regression grounds of data or information given to system.
(LR), XGBoost (XGB), Gradient Boosting Using information, model justifies a pattern using
Classifier (GBC) and Artificial Neural Network which the result is anticipated. They are addressed
(ANN). as Regression and Classification methods. Unaided
learning, while we perform unaided learning, the
A. Dateset information given to machine does not have any
labelling, thus machine has to make its own ways,
Dataset used has 76 attributed from them some are connecting various parameters [4]. It is used in
given [11]: circumstances in which we just need some
We have mentioned attributes with the details: connection between data or some hidden pattern
which are about impossible to find using human
1. Surveyed : Individual Identifier eyes, and it is all done without any human
2. Village : Village Identifier intervention, so chances of anomaly also decreases.
3.cons_nondurable: Non-durable expenditure As per our dataset, as dependent variable or result
(USD) variable we have has two possible conclusion,
4. asset_savings : asset_savings either the individual can be depressed or not
depressed. As we had labelled dataset, we had
5. cons_alcohol : Alcohol (USD)
applied classification calculation of directed
6. cons_tobacco : Tobacco (USD)
learning. Eight Distinct sort of arrangement
7. saved_mpesa: Saved money using M-Pesa calculation of machine learning is used .
8. durable_investment: Durable Investments 1. Decision tree (DT)
9. ed_expenses : Education expenditure past month 2. Random Forest (RF)
(USD) 3. Support Vector Machine (SVM)
10. early_survey : Psychology survey in 1st wave 4. Naïve Bayes (NB)
(dummy) 5. Logistic Regression (LR)
11. depressed: Meets epidemiological threshold for 6. XGBoost(XGB)
moderate depression (dummy) 7. Gradient Boosting Classifier (GBC)
8. Artificial Neural Network (ANN)
B. Data Pre-processing
A. Decision Tree (DT)
The process of converting some raw data into some
Decision tree is a concept which can be used in
meaningful information can be performed with
both, either classification or regression. In
some logical task and derive results [3]. This is
classification we mostly use decision tree, as it one
known as data pre-processing. For making a
of the most powerful and efficient tool [12].
machine learning model, this is one the most
Decision tree is a tree like structure classifier. After
crucial step, as in this phase we decide, if the
providing dataset as input in an algorithm, it makes
information is useful to us, or using which we can
a model out of it, further which we can use it to
have some insights.
predict results by providing our desired values or
In this process we do clean data, that means
information. The result will be anticipated on the
removing null values, assigning some values,
bases of the input dataset [6]. In Decision tree, each
derived from similar data or mean of data [13]. In
node represents a test on an attribute, outcome of
this phase we may also encounter some NA values,
the test is represented by the branch and leaf node
which we must remove, which we can do by
is treated as class label(terminal node) [17].
different ways [18]. Feature scaling is used to scale
Decision tree is drawn up side down, which mean it
different values to a same scale, so that we can
starts from the top, which means it’s root locates on
perform operations easily.
the top and leaf at bottom.
There was a column which was of date, we didn’t
In decision tree we have tree nodes:
want to make our prediction based on date, or make
date as a influencing factor, so we dropped the date 1. Chance Node – represented by circle
as column. Out of 86868 there were 13196 Nan 2. Decision Node – represented by square
values, which we removed and applied median 3. End Node – represented by triangle
function.
Construction of Decision tree
MODEL SELECTION Decision tree is made on the bases of outcomes
There are many strategies, which can be used to from the dependent attribute. For that we have to
find relevant results. There are two gatherings find entropy. Entropy is the possibility of getting a
possible from which we can characterize results of yes or no or we can say a 0 value or 1 value. To
those calculations, Regulated (Supervised) learning find the entropy of the class attribute, we have a
and unaided (Unsupervised) learning [14]. In formula:
Supervised learning, machine is prepared using
information. Then the result is anticipated on the

978-0-7381-4627-0/21/$31.00 ©2021 IEEE 1607

Authorized licensed use limited to: JAWAHARLAL NEHRU TECHNOLOGICAL UNIV. Downloaded on March 14,2024 at 10:03:23 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Third International Conference on Inventive Research in Computing Applications (ICIRCA-2021)
IEEE Xplore Part Number: CFP21N67-ART; ISBN: 978-0-7381-4627-0

Right(j)=child node from right split on node j”.

( ) ∑

Here E is the Entropy, s is the set and pi is the

frequentist probability of an element/class ‘i.

Fig.2 Random Forest Classifier (Wikimedia)

C. Support Vector Machine (SVM)

Support Vector machine Learning is a class of
machine learning which works on supervised
approach or on labelled data, in SVM the input data
have already all the required result. SVM predicts
good results in both classification and regression
problems. Using SVM we can separate 2 classes
Fig. 1 Decision tree Classifier points using a hyperplane. Center line between who
distinct values of which we have classified is
B. Random Forest (RF) known as hyperplane. SVM has its own unique
The only thing prevents tree to become ideal tool way of working as it just not only draw a
for predictive learning is inaccuracy. This means hyperplane, it also generates two margin lines,
trees work great while we use them with the data which will be having some distance, so that it could
which used to create them, but they are not flexible be easily linearly separable for each of the
enough. This is the reason, random forest is classification points. Margin planes will be linear to
considered better then decision as hyperplane, and they will be passing through the
It is adaptable, simple to use calculation, works nearest point we used to classify. Distance between
without hyper parameter tuning. It is easy to use margin plane and hyperplane is denoted using d+
and one of the most used algorithm. It is used as and d-. Distance between two margin plane is
both relapse and order assignment. known as Marginal distance. Using that Marginal
Working of Random Forest: distance, we can classify problem in much easier
Random Forest is a Supervised machine learning way. There colud be multiple hyperplane and
algorithm. Which means we have to use labelled marginal distance, but we are supposed to get the
dataset. Random Forest, as its name suggests, it marginal distance with maximum margin, so that
created random set of decision tree. Decision trees classification becomes easy. Support vectors are
are made on the bases of random attributes. the are points from which the margin line passes
Random sampling of training data points are used by.
when building trees. After getting different trees
with different attributes, all the trees are merged in
the end. That leads in increase in stability of the
model. As the decisions are made on the random
attribute, precision of result increases.
“For each decision tree Scit-learn calculates nodes
importance, using Gini Importance.
Mathematica formula for Gini importance:

( ) ( ) ( ) ( )
Fig.3 Support Vector Machine Classifier
It is considering 2 child nodes.
( ) ( )
Ni(j) = importance of node j
W(j)=Weighted number of samples reaching node j
Class labels are denoted as -1 for negative class and
C(j)= Impurity level of node j
Left(j)=child node from left split on node j +1 for positive class in SVM.

978-0-7381-4627-0/21/$31.00 ©2021 IEEE 1608

Authorized licensed use limited to: JAWAHARLAL NEHRU TECHNOLOGICAL UNIV. Downloaded on March 14,2024 at 10:03:23 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Third International Conference on Inventive Research in Computing Applications (ICIRCA-2021)
IEEE Xplore Part Number: CFP21N67-ART; ISBN: 978-0-7381-4627-0

in between 0 and 1, but never exactly at those

limits. Logistic Function makes a S type curve as
its resultant, which makes it very convenient
function. Using that S type curve, it becomes easy
The final optimization problem that we get fitting
to classify data.
the best parameters:
It is used as categorical data when the variable is
dependent. Like linear regression an equation is
used. Input values are combined linearly using
( ) weights to predict an output value. Key difference
between Linear and logistic regression is that, in
logistic regression, we get output value in binary on
D. Naïve Bayes (NB) the other hand in Linear regression we get output
value as numeric.
Naïve Bayes is a Supervised Machine learning In Logistic Regression, we use a cost function:
algorithm, Input data type is labelled. Naïve Bayes ( ( ))
algorithm is based on Bayes algorithm, or we can ( ( ) ) {
( ( ))
say that the Naïve Bayes algorithm is Advance
version or Bayes algorithm. Naïve Bayes works on
conditional probability, it means it a possibility is Cost function is used so that we can develop model
dependent on the probability on fulfilling of other with high precision and minimise error. Average of
events or the relationship with another event. loss function is considered as cost function. The
Working of Naïve Bayes: loss function is a value which is calculated at every
instance.
( ) ( )
( )
( )

Using this equation, we do find Naïve Bayes

probability of occurring A. Here A is the resultant
value and B is on which A is dependent. Using this
we can find probability of A(P(A)), given B is true.
B is termed as evidence. Probability of A is the
priori of A, which means probability of A even
before evidence is seen. Evidence is taken as an
attribute value of an unknown instance. Probability
of occurring Event A with respect to B(P(A/B)) is
a posteriori probability of B, that means possibility
of event after evidence is seen.
Fig.5 Logistic Regression Classifier (ODSC)

F. XGBoost

XGBoost is a gradient boosting framework-based

algorithm. It works somewhat similar to decision
tree.It can be used to predict various type of
information, which may be unstructured, such as
image, text. It is more efficient when it comes to
small to medium data size, and even can
outperform neural network. It was developed as
research project at the University of Washington
Fig.4 Naïve Bayes Classifier(cornell) [18]. Strong area of this algorithm is ranking,
regression and classification. It was developed to
E. Logistic Regression (LR) use resources such as time, computation power and
Logistic Regression is a Supervised machine memory efficiently. It have some different features
learning approach, which means we can use this in too, which makes it one of the unique algorithm.
both classification and regression. But Logistic Sparse Aware: Using this it can automatically
Regression is popular for its use in classification handle missing values.
[17]. It is used in Binary Classification. Logistic Block Structure: Using this many trees can be
Regression has taken from one of the Statistics generated simultaneously.
function, Logistic function also known as Sigmoid Continued training: Using this algorithm can
Function. Logistic Function is used to describe data further boost an already fitted model on new data.

978-0-7381-4627-0/21/$31.00 ©2021 IEEE 1609

Authorized licensed use limited to: JAWAHARLAL NEHRU TECHNOLOGICAL UNIV. Downloaded on March 14,2024 at 10:03:23 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Third International Conference on Inventive Research in Computing Applications (ICIRCA-2021)
IEEE Xplore Part Number: CFP21N67-ART; ISBN: 978-0-7381-4627-0

G. Gradient Boosting Classifier

Gradient Boosting is a one of the most powerful

technique to build predictive model. It is based on
the Adabost or adaptive boosting. It is based on the
concept of whether a weak learner, combined can
become better. The weak learners in AdaBoost are
decision trees with a single split, called decision
stumps for their shortness.
Working of Gradient Boosting:
Gradient Boosting uses three steps:
Loss Function: Loss function is dependent on how
we are using this algorithm. There are two types of
Fig.6 Human Neuron(lidolearning)
loss function, which are used by Gradient Boosting
Human’s neuron have dendrites which fetches data,
Algorithm, squared error, and logarithmic loss.
then there is a cell body, which helps in processing
While performing regression, squared error is used
the fetched data of dendrites, then there is Axom,
and for classification logarithmic loss is used.
using which we can pass our data to terminal axom
or synapse, which is further connected to another
Weak Learner : Decision tree, precisely regression
dendrites. By this way whole Neural network in
trees are used as weak learner. Using different trees
human is made. There are zillions of such neurons
we get output real values for splits and whose
using which our brain works. In machine too, we
output can be added together, allowing which using
try to make similar structure.
subsequent models output can be added and using
which we can make model more precise and near to
the actual value.

Additive Model: Using this Gradient Boosting, add

trees without touching the existing one. To
minimise the loss while adding tree, gradient
descent procedure is used. Gradient descent allows
algorithm to minimize the set of parameters. These
parameters include weights in neural network and
confident in a regression equation. After the loss is
Fig.7 Artificial Neuron (Vaibhav sahu)
calculated, while performing Gradient descent
procedure, to reduce loss we add we must add a How Artificial Neural Network Works:
tree. This approach is known as Functional descent In machine similarly we provide input as a dataset
or gradient descent with function.
to the neurons. Neural network is made up of
Layers of Neurons. First layer is known as Input
H. Artificial Neural Network (ANN) layer. Last layer is known as Output layer and in
between layers are known as hidden layers. Hidden
Neural Network term is derived from the term layers are responsible for most of the output we
Network of neuron. Base of this concept is Human receive. Firstly, data is transmitted to the first layer
brain, which have zillions of neurons, which can
of neurons. Neuron of first layer are connected to
work simultaneously, perform different action, and the next layer using channels [7]. Each channel is
capable of learning from outside using experiences. assigned some weight. Using weights, we decide
Artificial neural network is also supposed to work
the priority or the value of some input value. While
similarly, which have capacity to learn from processing First layer receives input and the input
experiences. Experience to the Artificial neural is multiplied by some weights (value of channel).
network is taught by providing dataset, using which And then the sum of all the product of input value
ANN prepares some model after finding some and weights are sent to hidden layer which have
common pattern between all the information some value known as bias. Then the value of bias is
provided. ANN is considered to one of the most added to the input sum [16]. This value then passed
efficient, reliable, and powerful approach towards to a threshold function known as activation
machine learning. Accuracy of ANN is much more function. That activation function decides if the
higher then the traditional approaches, because neuron will be activated or not. If the neuron is
ANN tries to learn using different aspects towards activated, then it will transmit data to the next
data, which helps in predicting better results then neuron using channels. This process is known as
other. Neurons in ANN is designed in similar way forward propagation. The output layer the neuron
as of humans. with highest value fires and determines the output.

978-0-7381-4627-0/21/$31.00 ©2021 IEEE 1610

Authorized licensed use limited to: JAWAHARLAL NEHRU TECHNOLOGICAL UNIV. Downloaded on March 14,2024 at 10:03:23 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Third International Conference on Inventive Research in Computing Applications (ICIRCA-2021)
IEEE Xplore Part Number: CFP21N67-ART; ISBN: 978-0-7381-4627-0

These are just probability. This prediction may be

right or wrong. Machine compares the given output Precision: This is the percentage of correct
with the actual output. Using this machine checks prediction. It defines the ratio of true positive to the
how much percent precited output is matched with sum of true positive and false positive.
the actual output.
Negative Predicted Value (NPV): It is a value
which is negative, and machine predicted the same.

F1 Score: Percentage of correct positive prediction.

False Negative Rate (FNR): It is a value, which is

true but stated false by machine.

False Positive Rate (FPR): It is a value Which was

initially false but stated true by machine.
Fig.8 Working of Artificial Neuron(Vaibhav
sahu) False Discovery Rate (FDR): While conducting
multiple comparison It is used to conceptualize type
This data then transmitted back to the network to 1 errors in null hypothesis.
reduce error. This is known as back propagation.
Using these information weights are adjusted. This False Omission Rate (FOR): It is a proportion for
forward and backward propagation performs using which true value is positive and predicted result is
various inputs. This happens until weights are negative.
adjusted correctly. This forward and backward
propagation is defined by epochs. Accuracy: It is a ratio of correctly predicted
observation to the total observation.
DATA ANALYSIS
Decision tree (DT)
Data analysis is one of the crucial stages of
Research. It is a somthing used by researchers to
reduce information to a story which is easy to
communicate and can be easily use to derive results
and insights [15]. It is a process which requires
statistical and logical techniques. After applying
some techniques we are supposed to evaluate or
analyse data, by which we can get some useful
information, suggesting conclusion, and supporting
decision making Table 1: Evaluation Parameters for Decision tree
There were two phase of experiment for this study:
(1) training phase, and (2) test phase. We divided Random Forest (RF)
data in 70/30 Ratio. Which means we used 70 % of
data in training and 30% in testing. The parameters
considered in the experiments were as follows: (1)
Sensitivity, (2) Specificity, (3) Precision, (4)
Negative Predicted Value (NPV), (5) F1 Score, (6)
False Negative Rate (FNR), (7) False Positive Rate
(FPR), (8) False Discovery Rate (FDR), (9) False
Omission Rate (FOR), (10) Accuracy
Table 2: Evaluation Parameters for Random Forest
PERFORMANCE AND RESULT ANALYSIS
Support Vector Machine (SVM)
Description of different Parameter considered to
evaluate the accuracy of model:

Sensitivity: Percentage of positive cases identified

correctly.

Specificity: This measures proportion of correctly

identified negatives

978-0-7381-4627-0/21/$31.00 ©2021 IEEE 1611

Authorized licensed use limited to: JAWAHARLAL NEHRU TECHNOLOGICAL UNIV. Downloaded on March 14,2024 at 10:03:23 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Third International Conference on Inventive Research in Computing Applications (ICIRCA-2021)
IEEE Xplore Part Number: CFP21N67-ART; ISBN: 978-0-7381-4627-0

Table 3: Evaluation Parameters for support Vector

Naïve Bayes (NB)

Table 8: Evaluation Parameters for Artificial Neural

Network

Table 4: Evaluation Parameters for Naïve Bayes Accuracy comparison of all the method applied

Logistic Regression (LR)

Table 5: Evaluation Parameters for Logistic Regression

XGBoost(XGB)
Bar Graph 1: Comparing results of all applied method

Table findings: We found that using various algorithms

Sensitivity, which means Percentage of positive cases
identified correctly was highest of Decision tree,
i.e., 87.87%. Highest Specificity, which measures
proportion of correctly identified negatives was of
Random Forest i.e., 50%. Highest Precision, that is
percentage of correct prediction. It defines the ratio
Table 6: Evaluation Parameters for XGBoost of true positive to the sum of true positive and false
positive was of both ANN as well SVM that is
Gradient Boosting Classifier (GBC) 100%. Highest Negative Predicted Value (NPV),
that is a value which is negative, and machine
predicted the same was highest of Naïve Byes, that
is 87.27%. Highest F1 score was of Gradient
boosting classifier, 91.81%. Lowest FNR was of
Decision tree,12.13%. Lowest FPR was of Random
Forest, that is 50%. Lowest FDR was of ANN and
SVM, that is 0%. Lowest FOR was of Naïve Bayes,
12.73%. And best accuracy overall was given by
SVM, that is 87.38%.

The dataset we used had 76 attributes. As we used

Table 7: Evaluation Parameters for Gradient Boost eight algorithms to fetch best results, the best we
Classifier
found was 87.38% and worst was 21.67%. Best
was using Support Vector Machine (SVM). The
Artificial Neural Network (ANN) worst result we got was using Naïve Bayes (NB).

These algorithm and procedure can be further used

to detect mental health of individual even before it
gets too late. The best result we got was 87.38%

978-0-7381-4627-0/21/$31.00 ©2021 IEEE 1612

Authorized licensed use limited to: JAWAHARLAL NEHRU TECHNOLOGICAL UNIV. Downloaded on March 14,2024 at 10:03:23 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Third International Conference on Inventive Research in Computing Applications (ICIRCA-2021)
IEEE Xplore Part Number: CFP21N67-ART; ISBN: 978-0-7381-4627-0

CONCLUSION Media." Proceedings of the ACM on Human-

Computer Interaction 3.CSCW (2019): 1-32
This paper shows a utilization of various Machine
Learning calculations for determining the mental [8] Graham, Sarah, et al. "Artificial intelligence for
health of an individual, using various common mental health and mental illnesses: an
know attribute, which are even available in numeric overview." Current psychiatry reports 21.11
values. This attribute contains earning and (2019): 1-18.
spending pattern of individual, also household
condition family members and investment was also [9] Bhattacharyya, R., & Basu, S. (2018). India Inc
an important part of the dataset. Which enabled us looks to deal with rising stress in employees.
to make such kind of model, which can predict Retrieved from ‘The Economic Times’.
mental health on the bases of income and
expenditure. [10] WHO. (2019, September 2). Suicide.
The examination of the outcomes connotes that the https://www.who.int/news-room/fact-
model can be used to get required information, sheets/detail/suicide
which is about 87.38 percent correct. Further this
research can be used in further improvement of the [11] b_depression. (2018, November 17). [Dataset].
system or making it more efficient. frankcc. https://www.kaggle.com/francispython/b-
depression.
References
[12] Shwetha, S, Sahil, A, Anant Kumar J, (2017)
[1] Reddy, U. Srinivasulu, Aditya Vivek Thota, and Predictive analysis using classification techniques
A. Dharun. "Machine learning techniques for stress in healthcare domain, International Journal of
prediction in working employees." 2018 IEEE Linguistics & Computing Research, ISSN: 2456-
International Conference on Computational 8848, Vol. I, Issue. I, June-2017
Intelligence and Computing Research (ICCIC) .
IEEE, 2018. [13] Tomar, D., & Agarwal, S. (2013). A survey on
Data Mining approaches for Healthcare.
[2] Shatte, Adrian BR, Delyse M. Hutchinson, and International Journal of Bio-Science and Bio-
Samantha J. Teague. "Machine learning in mental Technology, 5(5), 241-266.
health: a scoping review of methods and
applications." Psychological medicine 49.9 (2019): [14] Pang Z, Yuan H, Zhang Y-T, Packirisamy M.
1426-1448. Guest Editorial Health Engineering Driven by the
Industry 4.0 for Aging Society. IEEE J Biomed
[3] Van den Broeck, J., Cunningham, S. A., Heal Informatics. 2018;22(6):1709– 10.
Eeckels, R., & Herbst, K. (2005). Data cleaning:
detecting, diagnosing, and editing data [15] Pedregosa, F., Varoquaux, G., Gramfort, A.,
abnormalities. PLoS medicine, 2(10), e267. Michel, V., Thirion, B., Grisel, O., & Vanderplas,
J. (2011). Scikit-learn: Machine learning in Python.
[4] Mohr, David C., Mi Zhang, and Stephen M. Journal of machine learning research, 12(Oct),
Schueller. "Personal sensing: understanding mental 2825-2830.
health using ubiquitous sensors and machine
learning." Annual review of clinical psychology 13 [16] Bone, Daniel, et al. "Signal processing and
(2017): 23-47. machine learning for mental health research and
clinical applications [perspectives]." IEEE Signal
[5] American Psychological Association. (2012). Processing Magazine 34.5 (2017): 196-195.
Gender and stress.
https://www.apa.org/news/press/releases/stress/201 [17] T. Jain, V. K. Verma, M. Agarwal, A. Yadav
0/gender-stress and A. Jain, "Supervised Machine Learning
Approach For The Prediction of Breast Cancer,"
[6] Garcia-Ceja, Enrique, et al. "Mental health 2020 International Conference on System,
monitoring with multimodal sensing and machine Computation, Automation and Networking
learning: A survey." Pervasive and Mobile (ICSCAN), 2020, pp. 1-6, doi:
Computing 51 (2018): 1-26 10.1109/ICSCAN49426.2020.9262403.

[7] Chancellor, Stevie, Eric PS Baumer, and [18] Jain, Tarun, et al. "Performance Prediction for
Munmun De Choudhury. "Who is the" Human" in Crop Irrigation Using Different Machine Learning
Human-Centered Machine Learning: The Case of Approaches." Examining the Impact of Deep
Predicting Mental Health from Social Learning and IoT on Multi-Industry Applications.
IGI Global, 2021. 61-79.

978-0-7381-4627-0/21/$31.00 ©2021 IEEE 1613

Authorized licensed use limited to: JAWAHARLAL NEHRU TECHNOLOGICAL UNIV. Downloaded on March 14,2024 at 10:03:23 UTC from IEEE Xplore. Restrictions apply.