Rough-Logistics Regression Model Based on Purifying Inconsistent Information

Logistic regression (LR) is one of the widely applied methods for predicting categorical outcomes in medical research. Despite its popularity, the model is susceptible to data quality issues, particularly those related to outliers and inconsistent information. Outliers can only be identified using the characteristics of residual in the final stage, whereas inconsistent information can be addressed during preprocessing. This study focuses on enhancing binary LR by purifying inconsistent data through rough set theory. Two reduction strategies are examined: non-probability- and probability-based. The non-probability approach, the existing approach, eliminates all inconsistencies but substantially reduces sample size, which may weaken model reliability. To overcome this limitation, a probability approach is proposed, which allows inconsistent data to be transformed into consistent patterns while maintaining larger sample sizes. Empirical evaluation demonstrates that the probability-based strategy better preserves data variability, improves model robustness, and yields superior classification performance. Compared with conventional outlier treatments, the proposed method achieves higher accuracy, precision, sensitivity, and [Formula: see text]1-score while retaining more significant predictors. These results highlight the potential of RST-based data reduction, particularly the probability approach, as an effective preprocessing technique to strengthen LR analysis in medical datasets.