A Transportable Threshold-Based Framework for Interpretable Classification of Medical Data
A new framework uses Bernoulli Naïve Bayes with supervised chi-squared binarization to provide interpretable, rule-based clinical classification.
This approach addresses the black-box nature of medical AI by transforming continuous variables into interpretable thresholds. By maximizing associations with clinical outcomes, the method allows for transparent, rule-based decision-making without sacrificing the performance typically associated with continuous data models.
