Investigation of user fatigue based on input behavior

Abstract

Most of the computer users suffer from mental and physical fatigue. A tired person can make mistakes and violate sensitive data integrity or cause other problems in IT systems. Thus it is vital to keep the exhaustion of the user in check. In this research, we address the problem of identifying user fatigue level through the analysis of input behavior from the mouse and keyboard. Firstly, a specialized tool was developed, which was used to gather data about the keystroke dynamics and mouse motion characteristics. The following data were acquired from the keyboard: up-down, down-down, and holding time of the keys as well as keystroke frequency. Motion speed and hold time of the mouse key were gathered from the mouse. After the tool was created, a static text, as well as a combination of mouse inputs, were given for the volunteered users to make inputs. Corresponding data was gathered and labeled according to the user fatigue level. Neural network, K-Means as well as classification and regression tree algorithms were used to build user fatigue prediction models. Investigation on different datasets was performed, and the correlation between results obtained from keyboard and mouse datasets was analyzed. Analysis of the resulting accuracies of the models was performed as well, and corresponding conclusions about the capability of predicting user fatigue based on input behavior were made.