Human Activity Recognition in Prognosis of Depression Using Long Short-Term Memory Approach

Arnab Barua, Abdul Kadar Muhammad Masum, Erfanul Hoque Bahadur, Mohammad Robiul Alam, Md. Akib Uz Zaman Chowdhury,  Mohammed Shamsul Alam

Arnab Barua, Abdul Kadar Muhammad Masum, Erfanul Hoque Bahadur, Mohammad Robiul Alam, Md. Akib Uz Zaman Chowdhury, Mohammed Shamsul Alam

Abstract

Employments of Human Activity Recognition (HAR) have been introduced in contemplation of finding solutions to a handful of hazards associated with human health. Depression which can be introduced as one of the most mainstream mental disorder, has been found to have strong associations with human activity. Hence, HAR can be utilized to work out contemporary enactments so as to diminish the likelihood of suffering from depression with companion of appropriate measures. Our entire working arrangement is disposed into two segments. First segment encompasses the process of HAR exerting the sensors data from smartphones. Data from Accelerometer and Gyroscope sensor were put into service for conducting recognition of 13 human activities namely walking, walking upstairs, walking downstairs, sitting, standing, lying, jogging, cycling, sitting in toilet, fallen down, eating, drinking and irrelevant activities. The subsequent phase comprises computation of risk factor of suffering from depression by considering a person’s duration of conducting activities associated with depression. Employing a well-liked deep neural network namely Long Short-Term Memory (LSTM) on sensor data amassed from 10 subjects, we attained an accuracy about 95.85% on test data in accomplishment of HAR. In terms of computation of risk factor regarding depression, we elected 2 depressed subjects and 3 hearty subjects so as to evaluate our method’s performance. With employment of our method, we computed exorbitant risk factor of 67.44% and 74.92% from daily activities of those 2 depressed subjects and low risk factor of 29.86%, 27.91% and 29.87% in terms of mentioned normal subjects. Our method may come in handy to assist to control depression by accomplishment of the diminishment of rick factor of suffering from depression. Associations found between depression and human activities have obliged us to accomplish our method. In future, we are expecting a handful of advancement may be introduced in our method so as to bring our system in regular practice in our daily life.

Keywords: Accelerometer sensor; Depression; Human Activity recognition; Gyroscope sensor; Humidity sensor; Sensor; Temperature sensor; Long Short-Term Memory