Analysis of Machine Learning Algorithms for

Predicting Depression
MS. Purude Vaishali Narayanrao Dr. P. Lalitha Surya Kumari
Research Scholar, Professor,
Department of Computer Science And Engineering, Department of Computer Science And Engineering,
Koneru Lakshaiah University (KLEF), Koneru Lakshaiah University (KLEF),
Hyderabad, Telangana, India. Hyderabad, Telangana, India.

Abstract — The increasing use of technology can cause a individuals having different behavior. Prediction is
life- style of less physical work. Also, the constant pressure on performed using Predictive Analytics.
an individual can create a risk of mental disorder. These
vulnerabilities include peer pressure, heart attack, depression,
Along with statistics (both historical and current), Machine
and many other effects. In this paper different Learning algorithms, Predictive modeling and Data mining
approaches to predict depression are studied in detail. The techniques are used to estimate, or predict future outcomes.
mechanisms include collecting dataset through questionnaires Predictive Analytics is a branch of analytics that forecast any
asked to the person, posts on social media, text used future activity, behavior and trends which is called
throughout verbal communication and expressions on face. prediction.
Result is derived from extracted information. Here output Typically, this is the field of data scientists, statisticians and
expected is that the person needs attention or not. In skilled data analysts. Data relevant to this task is gathered by
this research work different algorithms and classifiers of Data engineers & then prepare it for analysis and also help
machine learning such as Decision Trees , SVM, Naive Bayes required for data visualization, dashboards and reports is
Classifier, Logistic Regression and KNN Classifier are
provided by software developers and business analysts. Data
analyzed to identify state of mental health in a target group.
Target groups used for this identification process are public
scientists use Predictive models to find correlations between
like students of high school, college students and working different data elements in stream data, patient health records
professionals. The paper also demonstrates an example in and other types of data sets. After collection of the dataset , a
which Twitter scrapping tool Twint is used detect whether statistical model is developed, trained and changed as
given Twit is depressive or not. required to get correct results. Against the selected data the
model is then run to generate predictions.
Index Terms—Machine Learning, Mental health, Predictive
III. RELATED WORK
Analytics.
Related work useful for this research is studied in detail.
I. INTRODUCTION In all studies different inputs are facial expression, Video,
As per WHO, 4.4% on globe are suffering with Text, Posts & comments on social media & behavioral
depression, among that varieties of depression are more features. Videos of population are captured & used as input
common in female than male. A major reason for suicide and for detecting depression [1]. Facial expressions from
global disability is Depression. If not treated, it affects day- captured video are identified. But expression changes
to-day life of people surrounding the person who is in depending upon pose, lighting, person, angle and sensors. So
depression, in family, at workplace or in society. Depression it’s a challenging task to encode features to detect
causes negative thinking, less concentration in work, reduced depression. Currently Deep learning architectures are
productivity. It also affects reproduction system of human providing good performance. For two dimensional & three
being. Depression may cause mental disorder also. Early dimensional Convolutional Neural Network & Recurrent
detection of mental disorder helps to start treatment at earlier
Neural Network, Spatial and temporal information is used. In
stage and improves quality of patient’s life as well as his
family. Mental health handles stress and is a measure for which a deep learning architecture is proposed to find the
taking decisions in each step in life. Mental Health is very ordinal relationship between the expressions on face and
important factor in every stage in life whether it be childhood intensity of depression based on distribution learning.
or an adult. Mental health could affect one’s thinking and Method depends on a new expectation loss.Thereby
behavior. Positive mental health will support person to have performance of the model is increased without the need to
full potential. It will also help him to cope up with the stress employ multiple channels. Performance is evaluated using
at home and workplace. It will increase the productivity of MAE and RMSE. When a video is inputted, score is
the people. calculated for each individual frame of video. All these
II. METHODOLOGY scores are averaged & based on it overall predicted
depression score is calculated. The average pooling
For predicting the Mental illness the algorithms of efficiently summarize spatio-temporal features. Hence author
Machine learning and Artificial Intelligence (AI) are very in [2] has detected that fully-connected layers are not
much useful. In real time these applications are helpful for required at all and it has also caused decrease in the number
the society as it will serve as a monitoring tool for of model parameters required along with over-fitting.

Authorized licensed use limited to: University College London. Downloaded on July 05,2020 at 00:19:04 UTC from IEEE Xplore. Restrictions apply.
In their experiments it is proved that when we combine possible ERDEo the prediction thresholds have been
chosen.
global & local responses, it shows increase in
performance and accuracy of C3D networks. While author In [4] a system based on AI is proposed which classifies
in [6] proposes a system that will help in detecting users according to their mental health. Input used for this
depression in college students. Features of happy, system is post made by user on Social Networking site
contempt and disgust faces are used to train the system. [SNS]. System scans user generated content (UGC) in the
post. Classifiers used are : SVM and Naïve Bayes. The
For testing phase college students are asked different
performance is analyzed using accuracy, precision and
questionnaires and videos of students are collected while recall. The accuracy means how much percentage of
they are answering different questionnaires. For effective depression is predicted correctly, the precision means how
detection of features throughout the video, the facial many predictions are positive (Non-Depressed), and the
features of these students are extracted and normalized. recall denotes the percent of the positive (Not- Depressed)
For the test dataset then the extraction of facial features correct cases caught. The experimental results are achieved
would be done, classifying them by SVM classifier for by testing 30 people. Out of 30 people, 15 are detected
depression detection. In the training phase the SVM depressive after answering the BDI-II questionnaire, and
classifier is trained with happy, contempt and disgust the remained are not predicted as depressed.
features extracted from the input happy face dataset. With In [8] previous work is reviewed and the overall phases of
this trained SVM the features extracted from each frame of MHMS(Mental Health Monitoring Systems) are presented
the video dataset can be tested to check if these features by proposing a Classification Taxonomy.
are present in each frame. An indication about the level of Different classification algorithms & clustering techniques
depression is obtained from the level of these features. are used to identify the mental health of an individual in a
Author in [14] has inputted Visual & Vocal information population. A prediction model is built for assessing the
using camera & microphone. These 2 inputs are used as mental health. In this research work clustering is used to
keys for developing strong system based on AI for identify the possible groups in the two target population
recognizing depression levels. taken under consideration. These are named as
population1 & population2. For each group 300 & 356
Using pertained CNNs artificial image processing can be samples were collected respectively. Then various
exploited by using the visual modality. For example clustering algorithms like K-Means clustering, hierarchical
clustering and partitioning around medoids or K- medoids
VGGFace provide good features at the frame level of
are applied. The techniques SVM,KNN, ensemble
videos. But it can only work for still images. Hence to (bagging) and tree ensemble (random forest) give an
capture spatial as well as dynamic temporal information equivalent accuracy score of 0.9. In order to validate the
technique called Feature dynamic history histogram labels obtained by clustering the concept of MOS are used.
(FDHH) technique is proposed. ML algorithms like PLS &
There is Correlation between depressive symptoms in
LR are used to map dynamic features and the depression
patients and behavioral features [17]. In this research work
scales. behavioral features are collected using mobile and
It is noticed that depression and other mental illnesses may wearable devices. For experimental result 46 persons are
lead to social withdrawal and isolation. As it is observed observed by author and identified 85 features which are
that social media is used by everyone to share happiness, unique and shows different behavior. This all data is
joy, achievements as well as sadness. Depressed people collected using 17 sensors. Then identified features were
will use social media for connecting with other people, for divided into 7 categories. Out of 85 some features are
sharing joy,happiness,sadness experience & for supporting found having major and consistent correlation that is
each other [3]. They are expecting professional help online directionality with mood assessment (e.g, the amount of
which can be provided using social media [13]. home stay, sleep duration, and vigorous activity), while
others showed directionality discrepancies across the
To identify & divide depression symptoms from written studies( e.g. amount of text messages [SMS] sent, time
texts in early stage, there are different ways based on spent between locations, and frequency of phone screen
machine learning methods. When a person is affected by activity).
depression the language person speaks will definitely
changes. Mental health can be monitored using count of
specific words by maintaining dictionary of such words. IV. EXAMPLE
The developed software is Linguistic Inquiry and Word As a example, Dataset collected from Data Driven Investor
Count (LIWC) [3]. For same purpose a Python Library (DDI) is used which is a small team with the mission to
Differential Language Analysis Toolkit (DLATK) was develop a guide from confusion to clarity in the age of
created which is open-source [3]. information, knowledge, and complexity[21]. It detects
depression and mental disorder using tweets from social
In [3] dataset is formed by continuously monitoring posts media Twitter by using algorithms of Machine Learning. It
& comments on a site reddit.com. It has a group of 752 used Baye’s classifier & python. The output is whether
users out of which 135 are depressed. Posts are searched person is depressed or not.
with a message which mention either depression or
treatment (e.g. “I undergone for treatment of depression”). Keywords are extracted from tweets using tool Twint and
If match is found then that user is identified as depressed depression is detected.
user. Author in this paper invented ERDE measure for
early detection of risk of depression. To represent the best

Authorized licensed use limited to: University College London. Downloaded on July 05,2020 at 00:19:04 UTC from IEEE Xplore. Restrictions apply.
A. Depressive Tweets V. CONCLUSION
As a cause of disability Depression has been
recognized in the whole world as it is causing more
considerable cost to health care systems. Depression leads
to negative thinking as well as insecure feeling. It affects
physical well-being and behaviors of the depressed person.
In more cases, depression is found as one of the leading
causes of suicide and substance abuse. In this paper
different approaches used for Predicting Depression are
analyzed and a Deep Learning mechanism is proposed for
automated depression detection. It allows to extract
features from collected dataset thorough text on Twitter.
This mechanism has potential to improve accuracy.

VI. REFERENCES
[1] Wheidima Carneiro de Melo, Eric Granger†, Abdenour
Hadid “Depression detection based on Deep Distribution
Learning” , ICIP 2019, Sep 2019, IEEE.
[2] Wheidima Carneiro de Melo, Eric Granger†, Abdenour
Hadid ”Combining Global and Local Convolutional 3D
Networks for Detecting Depression from Facial
Expressions” , IEEE FG, May 2019, IEEE.
[3]Marcel Trotzek, Sven Koitka, and Christoph M.
Friedrich “Utilizing Neural Networks and Lingustic
Metadata for Early detection of depression Indications in
Text Sequences” , IEEE TKDE, Dec 2018.
[4]Maryam Mohammed , Hafiz Farooq Ahmed
”Predicting depression levels using Social Media Posts” ,
Jun 2017,IEEE.
B. Positive Tweets [5]Sandhya P, Mahek Kantesaria “Prediction of Mental
disorder for employees in IT industry” , IJITEE Apr 2019.
[6]Dr. D. Venkataraman , Namboodiri Sandhya
Parameswaran “Extraction of Facial Features for
Depression Detection among Students”, 2018, IJPAM.
[7] Ms.Sumathi M.R., Dr. B. Poorna “Prediction of Mental
Health Problems among Children using Machine Learning
Techniques “, IJACSA 2016.
[8]Enrique Garcia-Ceja , Michael Riegler , Tine
Nordgreen ,Petter Jakobsen, Ketil J. Oedegaard , Jim
Tørresen “A Mental Health monitoring with Multimodal
Sensing & Machine Learning” , Science Direct, Sept.2018.
[9] Petra Hoffmannová “Text-Based Detection of the Risk
of Depression , Frontiers in Psychology”, March 2019.
[10]Anastasia Pampouchidou, Panagiotis Simos, Kostas
Marias,Fabrice Meriaudeau, Fan Yang, Matthew
Pediaditis, and Manolis Tsiknakis “Automatic Assessment
of Depression Based on Visual Cues: A Systematic
Review” ,IEEE Transactions on Affective
Computing,2017.
[11] M. Srividya “Behavioral Modeling for Mental Health
using Machine Learning Algorithms” , Springer Nature
2018.
[12]Amir Hossein Yazdavar, Hussein S. Al-Olimat,
Monireh Ebrahimi, Goonmeet Bajaj, Tanvi Banerjee,
Krishnaprasad Thirunarayan, Jyotishman Pathak, and
Amit Sheth “Semi-Supervised Approach to Monitoring
Clinical Depressive Symptoms in Social Media” , IEEE
Conference 2018 April 24.
[13] “A Framework for Classifying Online Mental Health-
Related Communities With an Interest in Depression” ,
IEEE J-BHI, VOL. 20, NO. 4, JULY 2016.
[14]Budhaditya Saha, Thin Nguyen, Dinh Phung, and
Svetha Venkatesh “Artificial Intelligent System for

Authorized licensed use limited to: University College London. Downloaded on July 05,2020 at 00:19:04 UTC from IEEE Xplore. Restrictions apply.
Automatic Depression Level Analysis Through Visual and
Vocal Expressions” , IEEE TCDS, VOL. 10, NO. 3,
SEPTEMBER 2018.
[15] Daniel Durstewitz1, Georgia Koppe, Andreas Meyer-
Lindenberg “Deep neural networks in psychiatry”,
Springer Nature Feb 2019.
[16]Asma Ghandeharioun, Szymon Fedor, Lisa
Sangermano, Dawn Ionescu, Jonathan Alpert, Chelsea
Dale, David Sontag, and Rosalind Picard “Objective
Assessment of Depressive Symptoms with Machine
Learning and Wearable Sensors Data” , 2017, ACII.
[17] Darius A Rohani, Maria Faurholt-Jepsen, Lars Vedel
Kessing, Jakob E Bardram “Correlations Between
Objective Behavioral Features Collected From Mobile and
Wearable Devices and Depressive Mood Symptoms in
Patients With Affective Disorders: Systematic
Review”,JMIR Mhealth Uhealth 2018.
[18] Khari, M., & Karar, A. (2013), “Analysis on intrusion
detection by machine learning techniques: a review”, Int.
J. Adv. Res. Computer. Sci. Software Eng, 3(4).
[19]Pritam, N., Khari, M., Kumar, R., Jha, S.,
Priyadarshini, I., Abdel-Basset, M., & Long, H. V. (2019).
“Assessment of code smell for predicting class change
proneness using machine learning”, IEEE Access, 7,
37414-37425.
[20]Khari, M., & Kumar, P. (2018). “Evolutionary
computation-based techniques over multiple data sets: an
empirical assessment”, Arabian Journal for Science and
Engineering, 43(8), 3875-3885.
[21] Viridiana Romero Martinez “A Machine Learning
Approach for the Detection of Depression and Mental
Illness in Twitter”, DDI, Feb 1, 2019

First Author is Asst. Prof. in the CSE Department of

Authorized licensed use limited to: University College London. Downloaded on July 05,2020 at 00:19:04 UTC from IEEE Xplore. Restrictions apply.