Data Augmentation of IMU Signals and Evaluation via a Semi-Supervised Classification of Driving Behavior

Over the past years, interest in classifying drivers' behavior from data has\nsurged. Such interest is particularly relevant for car insurance companies who,\ndue to privacy constraints, often only have access to data from Inertial\nMeasurement Units (IMU) or similar. In this paper, we present a semi-supervised\nlearning solution to classify portions of trips according to whether drivers\nare driving aggressively or normally based on such IMU data. Since the amount\nof labeled IMU data is limited and costly to generate, we utilize Recurrent\nConditional Generative Adversarial Networks (RCGAN) to generate more labeled\ndata. Our results show that, by utilizing RCGAN-generated labeled data, the\nclassification of the drivers is improved in 79% of the cases, compared to when\nthe drivers are classified with no generated data.\n