Previsão de atividades humanas para dispositivos móveis utilizando mineração de padrões sequenciais

Abstract

Smartphones have transformed the way people live and how they relate to others. As the number of sensors embedded in these devices increases, the amount of data that can be recognized grows, which makes these devices natural candidates for monitoring human activities. Therefore, there is a growing interest in the development of intelligent systems, such as virtual assistants, which use this information to recognize human behavior. One way to observe human behavior is by analyzing the routine of smartphone users. In computing, such identification can be done by using sequential pattern mining (SPM) techniques. The activities performed by the human being can be modeled in a symbolic sequence. From this sequence, SPM algorithms can be used to identify these patterns and, based on these, to identify which activity will occur in the future. However, such data are still little explored to predict future events. In this context, this work proposes a method for forecasting human activities recognized through mobile devices, based on sequential pattern mining. In addition, to improve the accuracy of the forecasting method, context data are used. These data have been increasingly explored and used in the field of intelligent systems and have shown promising results. The developed method is evaluated using two databases, reaching an accuracy of up to 84.40%. In addition, the results indicate that the use of context data in the scenario evaluated increases the accuracy of the method by up to 17.97%.