Scholarlink

Alzheimer’s disease is a neurodegenerative disorder characterized by irreversible damage, imposing a substantial burden on patients, their families, and society. Mild Cognitive Impairment (MCI) is widely regarded as a prodromal stage of Alzheimer’s disease. Timely detection of MCI is crucial for the prevention of Alzheimer’s disease and halting further progression. Exploring potential factors influencing MCI onset and automating MCI prediction through screening scales and assessments of daily living status in the elderly enable low-cost, rapid, and high-accuracy identification of MCI. This study focuses on leveraging machine learning techniques to screen for and predict the occurrence of MCI using readily accessible information sources, including demographic data, psychological health measures, and life history interviews. Furthermore, it aims to identify key factors influencing MCI risk based on the trained predictive models. Given the scarcity of comprehensive multifactorial survey data on MCI within the Chinese population, this research employed a self-designed interview protocol and questionnaire to investigate cognitive function and associated factors in individuals aged 60 years and above. Machine learning algorithms including XGBoost, Random Forest, and Logistic Regression were fitted to model the data. Among these, the XGBoost classifier achieved the highest accuracy rates of 0.89 on the test sets, respectively. SHapley Additive exPlanations (SHAP) were utilized to interpret the model’s decision-making process and analyze the contribution of various factors to MCI prediction. Results identified depression and age as two primary influencing factors in MCI prediction and analysis. This highlights the importance of focusing on these factors in understanding the development of MCI.

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

SHAP-Based Interpretion of Machine Learning Prediction Models for Mild Cognitive Impairment

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Zongqin Wang,
  • Zhong Ding,
  • Chenling Liu,
  • Baoliang Zhong,
  • Jin Lu

摘要

Alzheimer’s disease is a neurodegenerative disorder characterized by irreversible damage, imposing a substantial burden on patients, their families, and society. Mild Cognitive Impairment (MCI) is widely regarded as a prodromal stage of Alzheimer’s disease. Timely detection of MCI is crucial for the prevention of Alzheimer’s disease and halting further progression. Exploring potential factors influencing MCI onset and automating MCI prediction through screening scales and assessments of daily living status in the elderly enable low-cost, rapid, and high-accuracy identification of MCI. This study focuses on leveraging machine learning techniques to screen for and predict the occurrence of MCI using readily accessible information sources, including demographic data, psychological health measures, and life history interviews. Furthermore, it aims to identify key factors influencing MCI risk based on the trained predictive models. Given the scarcity of comprehensive multifactorial survey data on MCI within the Chinese population, this research employed a self-designed interview protocol and questionnaire to investigate cognitive function and associated factors in individuals aged 60 years and above. Machine learning algorithms including XGBoost, Random Forest, and Logistic Regression were fitted to model the data. Among these, the XGBoost classifier achieved the highest accuracy rates of 0.89 on the test sets, respectively. SHapley Additive exPlanations (SHAP) were utilized to interpret the model’s decision-making process and analyze the contribution of various factors to MCI prediction. Results identified depression and age as two primary influencing factors in MCI prediction and analysis. This highlights the importance of focusing on these factors in understanding the development of MCI.