Scholarlink

This study integrates ensemble machine learning models and explainable artificial intelligence (XAI) frameworks for disease classification, focusing on anemia and malaria. For anemia classification, data is sourced from the South African National Health and Nutrition Examination Survey (SANHANES-1), involving 493 individuals aged 15–17 years (264 females and 229 males). Five different ensemble methods were used, including random forest, AdaBoost, gradient boost, XGBoost, and CatBoost. This study dealt with class imbalance by oversampling technique, therefore improving the robustness of the predictions. From the ensemble models, CatBoost stood out as the best with ROC AUC score of 0.980, followed closely by random forest at 0.973, while gradient boost, XGBoost, and AdaBoost achieved scores of 0.963, 0.964, and 0.900, respectively. On the other hand, malaria classification that used secondary data from the Federal Polytechnic Ilaro Medical Centre, Ilaro, Ogun State, Nigeria, comprising information from 337 patients aged between 3 and 77 years (180 females and 157 males) over a 4-week period, was utilized. Ensemble methods, including random forest, AdaBoost, gradient boost, XGBoost, and CatBoost, were employed, addressing class imbalance through oversampling technique. From the ensemble models, random forest stood out as the best with an ROC AUC score of 0.840, followed closely by CatBoost with an ROC AUC score of 0.795. XGBoost, gradient boost, and AdaBoost achieved ROC AUC scores of 0.790, 0.747, and 0.651, respectively. Explainable AI techniques such as LIME and permutation feature importance were utilized to enhance transparency and interpretability in both studies, evaluating how important different characteristics might influence the probability of getting the disease. These methodologies reveal which characteristics are essential for making predictions, hence promoting transparency and comprehension of how models make decisions. This information is necessary for healthcare providers because it may lead to better treatment options and improved patient outcomes, particularly concerning anemia and malaria management.

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

Explainable AI Models for Improved Disease Prediction

  • 章节

  • 发表日期：

    2025年

查看全文
  • Peter Mwangi,
  • Samuel Kotva,
  • O. Olawale Awe

摘要

This study integrates ensemble machine learning models and explainable artificial intelligence (XAI) frameworks for disease classification, focusing on anemia and malaria. For anemia classification, data is sourced from the South African National Health and Nutrition Examination Survey (SANHANES-1), involving 493 individuals aged 15–17 years (264 females and 229 males). Five different ensemble methods were used, including random forest, AdaBoost, gradient boost, XGBoost, and CatBoost. This study dealt with class imbalance by oversampling technique, therefore improving the robustness of the predictions. From the ensemble models, CatBoost stood out as the best with ROC AUC score of 0.980, followed closely by random forest at 0.973, while gradient boost, XGBoost, and AdaBoost achieved scores of 0.963, 0.964, and 0.900, respectively. On the other hand, malaria classification that used secondary data from the Federal Polytechnic Ilaro Medical Centre, Ilaro, Ogun State, Nigeria, comprising information from 337 patients aged between 3 and 77 years (180 females and 157 males) over a 4-week period, was utilized. Ensemble methods, including random forest, AdaBoost, gradient boost, XGBoost, and CatBoost, were employed, addressing class imbalance through oversampling technique. From the ensemble models, random forest stood out as the best with an ROC AUC score of 0.840, followed closely by CatBoost with an ROC AUC score of 0.795. XGBoost, gradient boost, and AdaBoost achieved ROC AUC scores of 0.790, 0.747, and 0.651, respectively. Explainable AI techniques such as LIME and permutation feature importance were utilized to enhance transparency and interpretability in both studies, evaluating how important different characteristics might influence the probability of getting the disease. These methodologies reveal which characteristics are essential for making predictions, hence promoting transparency and comprehension of how models make decisions. This information is necessary for healthcare providers because it may lead to better treatment options and improved patient outcomes, particularly concerning anemia and malaria management.