Scholarlink

<p>Explainable Artificial Intelligence (XAI) plays a pivotal role in enhancing the transparency and trustworthiness of machine learning models, particularly in domains where human understanding is crucial. This study presents a comparative evaluation of white-box and black-box classifiers applied to an educational dataset, emphasizing both predictive performance and interpretability. Among the white-box models, Logistic Regression achieved strong overall performance (95% accuracy, F1-scores of 0.93 and 0.96), closely followed by K-Nearest Neighbors (89%), Decision Tree (88%), and Naive Bayes (87%). Similarly, black-box models such as Random Forest (91%), Multilayer Perceptron (89%), Gradient Boosting (88%), and Support Vector Machine (87%) delivered competitive results. To strengthen the reliability of the findings, the models are further validated using an external dataset, where the performance trends remained consistent, confirming the robustness and generalizability of the evaluated classifiers. While the performance differences across models are marginal and not statistically significant, the interpretability advantage of white-box approaches particularly Logistic Regression highlights their suitability in educational contexts where model transparency and explainability are essential alongside accuracy. These findings underscore the importance of balancing predictive performance with interpretability in human-centered AI applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Interpretable learning models: an XAI-focused evaluation of classifier performance

  • 期刊论文

  • 发表日期：

    2026年3月13日

查看全文
  • Pratiyush Guleria

摘要

Explainable Artificial Intelligence (XAI) plays a pivotal role in enhancing the transparency and trustworthiness of machine learning models, particularly in domains where human understanding is crucial. This study presents a comparative evaluation of white-box and black-box classifiers applied to an educational dataset, emphasizing both predictive performance and interpretability. Among the white-box models, Logistic Regression achieved strong overall performance (95% accuracy, F1-scores of 0.93 and 0.96), closely followed by K-Nearest Neighbors (89%), Decision Tree (88%), and Naive Bayes (87%). Similarly, black-box models such as Random Forest (91%), Multilayer Perceptron (89%), Gradient Boosting (88%), and Support Vector Machine (87%) delivered competitive results. To strengthen the reliability of the findings, the models are further validated using an external dataset, where the performance trends remained consistent, confirming the robustness and generalizability of the evaluated classifiers. While the performance differences across models are marginal and not statistically significant, the interpretability advantage of white-box approaches particularly Logistic Regression highlights their suitability in educational contexts where model transparency and explainability are essential alongside accuracy. These findings underscore the importance of balancing predictive performance with interpretability in human-centered AI applications.