International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 08 Issue: 04 | Apr 2021 www.irjet.net p-ISSN: 2395-0072 © 2021, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 3449 Methodologies for Depression Detection using Smart Wearables Ritom Gupta 1 , Pranav Bakre 2 , Pratik Gorade 3 , Vignesh Iyer 4 , Shamla Mantri 5 1-4 Final Year, B.Tech CSE, Dr. Vishwanath Karad MIT World Peace University, Pune, India 5 Professor, Dr. Vishwanath Karad MIT World Peace University, Pune, India -------------------------------------------------------------------------***---------------------------------------------------------------------- Abstract - Detecting depression using wearable smart devices is a rapidly emerging method in the detection and diagnosis of depressive disorders. The previous in this domain have used behavioural techniques, machine learning and correlation analysis to detect/diagnose depression. In this work, we recapitulate the methods of data collection, self-reporting mechanisms, decision methods and parameter correlation with depression tendencies. This work aims to give a direction for future research in this domain, by identifying the needs and concerns to be worked upon. We have identified the need for greater emphasis on intraindividual variability in mood, multimodal approaches, and general solutions. Keywords: Machine Learning, Correlation, Ecological Momentary Assessment, Depression Scale, DSM, HAMD, Multimodal Techniques 1. INTRODUCTION This work is an in-depth review of depression detection techniques using smart wearables. Prominence is given to various approaches which establish correlation of parameters with depression. This work aims to throw some light on the existing works with the main questions being: 1) What are the parameters closely related to depression? 2) What are the most used techniques? 3) Which methods have been used to collect data? 4) What are the areas to further examine? A thorough Internet search was done using key words like depression, fitness data, smart wearables. This work encapsulates the works based on quantitative analysis, advantages and limitations. Data collection protocols and the corresponding methods applied to analyse presence/severity of depression have been described. 2. RELATED WORK Numerous studies have been carried out for predicting depression and measuring its severity using smartphones and wearables. Most of the studies have found a relation between predicted severity of and actual severity. The various mechanisms used so far have ranged from simple questionnaires and passive sensing to machine learning algorithms, EEG-based measurement and some combinations of these methods allied together in developing a system. In person submission of recorded data in the organizer’s lab after a period of having the wearable collecting their data has been a common method for collecting data. In most cases, data has been collected from specific groups like children, university students, unemployed men, women, etc. Number of people whose data was collected is between 10-100, possibly, due to the human effort involved in collection, limited funding and the time required to get enough data for thorough scanning for one person. The sensitivity of clinical data being high, prevents data from being made public. With the known mentioned practices in collection, we go through the related works in this field. After gathering the answers to 11 question survey from the stakeholders (healthcare providers, people having or previously diagnosed with depression and their caregivers, healthcare insurance companies, etc), the results showed that majority of the stakeholders somewhat or strongly agree that the music streaming service can be used as an addition in therapy for depression. Music streaming accentuated as an additive in digital therapy for depression [1]. In another approach, 40 participants were categorized as depressed, their GPS data and phone usage data were monitored for 2 weeks which was used to derive 10 different parameters, namely: Location variance, Number of location clusters (K), Entropy, normalized entropy, home stay, circadian movement, total distance, phone usage frequency, phone usage duration. Linear regression was used to estimate the PHQ-9 score and actual PHQ-9 scores were used as a test. Logistic regressor classification was used to classify participants as depressed/not depressed. Significant correlation of variables was discovered between simplified entropy, varying location, staying home, and the PHQ-9 scores. Mobile usage data suggests correlation with parameters like usage duration, and usage frequency (Correlation Coefficient =.54, P- value=.011, and Correlation coefficient=.52, P-value=.015). An accuracy of 86.5% was achieved using the normalized