Adaptive MSD-Splitting Technique Enhances Decision Tree Algorithms for Skewed Data

A novel computational technique known as Adaptive MSD-Splitting (AMSD) has been introduced to enhance decision tree algorithms for continuous numerical attributes. This new method overcomes the shortcomings of the traditional MSD-Splitting, which relies on empirical mean and standard deviation to optimize the C4.5 algorithm's performance. While conventional MSD-Splitting is effective for symmetric distributions, its rigid one-standard-deviation thresholds can lead to considerable information loss in highly skewed datasets typical in biomedical and financial fields. AMSD adjusts the standard deviation multiplier according to feature skewness, refining intervals in dense areas to preserve discriminative power. This approach has been incorporated into ensemble methods, particularly improving Random Forests. The findings build on prior research in arXiv:2604.19722v1, highlighting the challenge of discretizing continuous attributes in decision tree induction as dataset dimensions grow. The announcement was shared via a cross-type abstract on the arXiv platform, showcasing the technical nature of this advancement.