Integration of DEA-Based Inefficiency into Decision Tree Splitting Criteria

Abstract

This paper introduces a modified decision tree methodology that incorporates efficiency evaluation through Data Envelopment Analysis (DEA). Standard decision trees rely on class distribution metrics such as the Gini index or Entropy, overlooking the performance characteristics of individual observations. In contrast, the proposed approach calculates inefficiency scores for each Decision-Making Unit (DMU) based on input excesses and output shortfalls. It uses these scores to weight samples during tree construction. This integration allows the model to achieve accurate classification while simultaneously highlighting units with poor performance. Results indicate that the DEA-informed decision tree effectively captures inefficiency patterns and provides actionable insights for performance improvement, while maintaining predictive accuracy.