Associative spatio-temporal classification: a scalable spatio-temporal classifier

The analysis of data characterized by information regarding the location and the time of events presents several challenges. These challenges are related to finding meaningful representations for the data and being able to process them in massive quantities. When dealing with the classification of spatio-temporal data, another open issue is the capability of identifying and predicting rare critical events. The objective of this thesis is to present a novel associative classifier to tackle these problems. The classifier was trained and tested on a public dataset of bike-sharing data in the San Francisco Bay area over a two-year-long period. The model was built with Python and Spark, and three other types of classifiers were built to assess its performance. The results show a superior performance in terms of precision and better resilience to missing values. The results highlight the importance of an effective data representation of the spatio-temporal events, and that interpretable models can provide good insights to enhance the performance of prediction algorithms, even in cases where the interpretability has not a crucial importance.