A Method and System for Predicting Explainable Business Process Outcomes Based on Rule-Based Neural Networks

Abstract

This invention belongs to the interdisciplinary field of artificial intelligence and business process management, and discloses a method and system for predicting interpretable business process results based on rule-based neural networks. This invention converts business process instance trajectories into binary feature vectors based on declarative constraints and groups them according to semantic type. Through the TemporalDRNet model, its inter-group interaction rule generation layer selects features from each group based on dynamic biases and generates rules simulating the logical "AND". The temporal extraction aggregation layer aggregates rules with non-negative weights and simulates the logical "OR" to output the prediction. The trained model weights can be directly mapped to the IF-THEN business rule set that explains the prediction results. The end-to-end deep neural network model TemporalDRNet proposed in this invention achieves multi-dimensional technological breakthroughs through innovative designs such as Declare constraint fusion, semantic grouping modeling, and end-to-end binarization optimization.

Images