ABSTRACT
Stress varies from person to person. It is challenging to measure an individual’s stress because of its personalized nature. But with the help of smartphones, stress level detection can be made easy because almost all smartphones are made up of inbuilt sensors. We have used these sensors to identify stress levels. Data was collected from 51 users for 6 months using the “Sensors Recorder” app. It collects data from accelerometer, gyroscope, and ambient light sensors as well as call data and app usage in the background. In the foreground, it collects the stress level of the user by giving a daily notification. After the data extraction, Exploratory Data Analysis is performed to know the correlation between sensor data features and stress levels. A Long Short Term Memory Model is used for the prediction of stress levels. Five models were trained on different permutations of the extracted features and the best model gave an accuracy of 82.81%.
