Tree-based Ensemble Learning for Stress Detection by Typing Behavior on Smartphones

Sukma Darmawan, A.; Ernawan, F.; Benawan, I.; Akbar Andriawan, Z.; Wibowo, A.; Sugiharto, A.; Adi Sarwoko, E.; Kusuma, M.

Stress is an emotional feeling that arises in a person in response to unpleasant external pressures or demands. Stress can be obtained from environmental, work, academic, economic, and other problems. The COVID-19 pandemic, which is changing people's lifestyles, has a major impact on people's mental health. The experiments used 1522 respondents that related to stress problems due to the COVID-19 pandemic from the Indonesian Association of Mental Medicine Specialists (PSDKJI), the prevalence reached 64.3%. There are many different ways to detect stress, but most of them require special tools to identify it. This paper presents a prediction based on machine learning for identifying the stress levels by typing behavior data on a smartphone keyboard. This study uses tree-based ensemble learning to make predictions. Based on the experimental results, the Random Forest model produces the best accuracy with a value of 94.77%. The most influential feature of the model is the standard deviation feature of the gravity sensor on the Z-axis. © 2021 IEEE.