Mining for spatio-temporal distribution rules of illegal dumping from large dataset

Illegal dumping has been an issue to be dealt with by the authorities. The incidents distribute across spatial and temporal domains, possibly with recurring patterns. To assist in addressing the issue, these patterns in the form of classification rules can potentially be mined from large datasets collected by the authorities. This research represents a novel work in discovering rules described by spatio-temporal features of the illegal dumping activities. A feature selection methodology that considers a range of techniques employing differing optimality criteria is proposed. A hybrid algorithm is developed by combining the proposed method to the C4.5 algorithm. A series of experiments demonstrated the advantages of the proposed algorithm. The feature selection approach is shown to balance the different optimality criteria, overcoming the dominance of any individual criteria. This work further shows that the generated spatio-temporal rules, when generated and implemented in information systems, are potentially applicable in preventive and enforcement work by the authorities.