An approach for the structural learning of chronicles

Abstract

The Chronicles are patterns characterized by observable events, with temporal relationships between them. In this work, we propose a model for the building of chronicles, through a learning strategy that allows defining its structure. Our approach discovers the events that compose a chronicle and the temporal relationships between them. These events are defined by changes in the descriptors/features of the modeled phenomena, according to the sequence in which they appear. We test our approach for modeling a hierarchical pattern of vehicle driving styles, which consists of three levels, one to describe the emotional states, another to describe the driver states and, finally, the last one to describe the driving styles. Finally, our approach is compared with other techniques, in classical benchmark classification problems.

Recognition of the Driving Style in Vehicle Drivers

“…In general, the chronicles can describe the same situation (an emotion, a driver state, a driving style) using different chronicles. This approach requires a robust chronicle database, which is constantly learned, in order to adapt it to the driver and new situations [30,43]. With respect to Ar2p, it uses two strategies of adaptation [34].…”