Diagnosis and Prediction Model for Alzheimer's Disease based on Random Forest Algorithm

Abstract

Early diagnosis of Alzheimer's disease (AD) is crucial for delaying disease progression. Although machine learning methods such as random forests have shown potential in this field, effectively leveraging multi-source clinical data to construct high-accuracy models remains a research priority. This study aims to develop and validate a diagnostic and predictive model for AD based on the random forest algorithm. Using a multidimensional dataset encompassing demographic information, clinical test indicators, and lifestyle factors, the study first performed data cleaning and feature engineering, innovatively constructing composite features such as cardiovascular risk scores, cognitive impairment risk scores, and comprehensive health scores. Subsequently, a feature selection method incorporating domain knowledge was applied to identify a core set of features, including Activities of Daily Living (ADL), functional assessments, Mini-Mental State Examination (MMSE), and age. The final random forest model, optimized through cross-validation, achieved an accuracy of 94.95% and an F1-score of 0.9476 on the test set. Feature importance analysis not only validated the model’s interpretability but also highlighted the critical predictive roles of patients’ functional status and cognitive levels. The findings demonstrate that the random forest-based model can efficiently and accurately support the auxiliary diagnosis of AD, offering a low-cost, non-invasive solution with potential applications for early screening in primary healthcare settings.