Classification of Soft Keyboard Typing Behaviors Using Mobile Device Sensors with Machine Learning

The amount of personal data stored on mobile devices has risen significantly during the past several years as a result of two developments: More people are using them, and sensors have become more advanced, capable of analyzing and classifying human activities such as walking, running, sleeping and cycling, and swimming. In this study, we propose a system to classify users' typing behaviors based on the data produced by the built-in sensors and present a login use case scenario to validate the results. We investigate users' unique typing and phone holding behaviors by examining the soft biometric (age, gender) and statistical features. Typing behaviors are classified by various machine learning techniques with the data inputted from accelerometer and gyroscope sensors. Artificial neural networks (ANN), k-nearest neighbors (k-NN), support vector machines (SVM) and RandomForest Classifier (RFC) algorithms, which are some of the most common algorithms, were applied for classification. In the user studies, we achieved accuracy of 98.55% for ANN, 100% for k-NN, 99.8% for SVM and 99.5% for RFC. The system is capable of device-based training and can distinguish the device owner's typing behavior from those of others with 100% accuracy. The proposed system was tested on a developed mobile application prototype, and its applicability was shown through experiments.