<p>This paper presents a monotonic decision tree framework to tackle ordinal classification challenges by integrating correlation screening, rank mutual information, and monotonicity constraints. The proposed monotonic decision tree introduces a two-stage screening mechanism under the principle of “threshold admission, information first” to select the optimal splitting feature. In the first stage, the Kendall’s rank correlation coefficient is utilized to screen for candidate feature set satisfying the monotonicity threshold. In the second stage, we obtain the final splitting feature by combining the Kendall’s rank correlation coefficient and rank mutual information. The experimental results demonstrate that the proposed monotonic decision tree is feasible and effective in terms of model compactness, ordinal prediction accuracy, noise robustness, and construction efficiency.</p>