Transportation Mode Characterization; Information Theory; Causal Planes;
Information Theory, human mobility e ML
The uncontrolled growth caused by migration and the increasing world population demand solutions to improve the infrastructure of cities. The academic and industrial communities are investing in solutions centered on human mobility to improve the life quality of humans. In this context, location-aware services provide valuable information for capturing human mobility patterns. Exploring the dynamics of mobility time series leads to better solutions by understanding the underlying data-generating process and identifying different patterns. Strategies based on Information Theory quantifiers associated with Ordinal Pattern (OP) methods, like Complex-Entropy Causality Plane (CECP) and Fisher-Shannon Causality Plane (FSCP), have reached relevant advancements in distinguishing different time series dynamics. Therefore, they are promising tools to explain those complex behaviors to improve human mobility-based services. Based on that, this thesis presents an approach to characterize and classify transportation modes through the Information Theory quantifiers computed over their speed time series. This approach presents high-quality data preprocessing to ensure reliability and validity. We characterize each transportation by examining its mobility aspect changes over time and comparing its statistical properties to noises mapped onto the causal planes. Finally, we build and evaluate robust Machine Learning classifiers centered on Information Theory quantifiers computed over the time series. Evaluation results show the potential of our study, allowing us to identify motorized and non-motorized means of transportation regimes, estimate transportation switching based on causal plane mappings, and efficiently classify transportation categories through machine learning models.