Instance-Based Discovery of ItemSBs Leveraging Individuality Through Discretization

This study proposes a novel instance-based method for discovering itemsets with statistically distinctive backgrounds (ItemSBs) by discretizing continuous attribute values. Instead of discretizing an entire dataset, in this method, each instance's continuous attributes are discretized individually using Standard Deviation-Based Discretization and Segmented Distance-Order-Based Discretization techniques, capturing the instance-specific characteristics of ItemSBs. The proposed method enables the discovery of explainable ItemSBs and forecasting by leveraging the machine rule mining method GNMiner. Experiments performed on a music dataset demonstrated the effectiveness of the proposed approach in improving the discovery of ItemSBs and the associated forecasting accuracy compared with conventional methods that overlook individuality.