Predicting Useful Information From Typing Patterns Using a Bootstrapped-Based Homogeneous Ensemble Approach

Nowadays, the way a user connects with computing devices is being analysed with the goal of extracting useful information for some interesting applications beyond user authentication. It enables the development of next-generation intelligent human-computer interaction, auto-profiling, and soft biometrics. In this paper, a bootstrapped-based homogeneous ensemble model has been proposed without wasting rare samples in each bootstrapped training set to overcome the uneven distribution of classes (common in keystroke dynamics) for predicting users’ traits, fine motor skills, and cognitive deficiency automatically using the user’s daily typing habit. This model is lightweight, faster, and could be implemented in low-configured devices like smartphones. The proposed model has been verified with a more realistic evaluation and several shared and authentic keystroke dynamics (KD) datasets and achieved 93.21% of accuracy in predicting age group, 65.35% in identifying gender, 87.14% for handedness, 77.14% in the case of hand(s) used determination, 91.25% in predicting qualification, 74.44% in recognising typing skill, 58.45% in observing lie, 84.12% in the determination of Parkinson’s disease (PD), and 99.14% in predicting emotional stress (ES). The proposed model might be used for a wide range of more interesting applications, including automatic user profiling in social networking, age-restricted access control to protect kids from Internet threats, age- and gender-specific product recommendations in e-commerce, medical diagnostics at-home environment for better treatment and therapy management, soft biometric traits in improving biometric models, learner’s cognitive deficiency in effective online teaching and learning, cognitive deficiency marker in a competitive examination, unbiased online feedback collection, and online inquiry correctness measurement, to mention a few.