Sparse episode identification in environmental datasets: The case of air quality assessment

Sparse episode identification in environmental datasets is not only a multi-faceted and computationally challenging problem for machine learning algorithms, but also a difficult task for human-decision makers: the strict regulatory framework, in combination with the public demand for better information services, poses the need for robust, efficient and, more importantly, understandable forecasting models. Additionally, these models need to provide decision-makers with “summarized” and valuable knowledge, that has to be subjected to a thorough evaluation procedure, easily translated to services and/or actions in actual decision making situations, and integratable with existing Environmental Management Systems (EMSs).On this basis, our current study investigates the potential of various machine learning algorithms as tools for air quality (AQ) episode forecasting and assesses them – given the corresponding domain-specific requirements – using an evaluation procedure, tailored to the task at hand. Among the algorithms employed in the experimental phase, our main focus is on ZCS-DM, an evolutionary rule-induction algorithm specifically designed to tackle this class of problems – that is classification problems with skewed class distributions, where cost-sensitive model building is required.Overall, we consider this investigation successful, in terms of its aforementioned goals and constraints: obtained experimental results reveal the potential of rule-based algorithms for urban AQ forecasting, and point towards ZCS-DM as the most suitable algorithm for the target domain, providing the best trade-off between model performance and understandability.