The global increase in dementia cases highlights the importance of early detection and intervention, particularly for individuals at risk of mild cognitive impairment (MCI), a precursor to dementia. The aim of this study is to develop and validate machine learning (ML) models based on non-imaging features to predict the risk of MCI conversion in cognitively healthy older adults over a three-year period. Using data from 845 participants aged 65 to 87 years, we built five eXtreme Gradient Boosting (XGBoost) models of increasing complexity, incorporating demographic, self-reported, medical, and cognitive variables. The models were trained and evaluated using robust preprocessing techniques, including multiple imputation for missing data, Synthetic Minority Oversampling Technique (SMOTE) for class balancing, and SHapley Additive exPlanations (SHAP) for interpretability. Model performance improved with the inclusion of cognitive assessments, with the most comprehensive model (Model 5) achieving the highest accuracy (86%) and area under the curve (AUC = 0.8359). Feature importance analysis revealed that variables such as memory tests, depressive symptoms, and age were significant predictors of MCI conversion. In addition, an online risk calculator has been developed and made available free of charge to facilitate clinical use and provide a practical, cost-effective tool for early detection in diverse healthcare settings ( https://aimar-project.shinyapps.io/MCI-risk-calculator/ ). This study highlights the potential of non-imaging ML models for early detection of MCI and emphasizes their accessibility and clinical utility. Future research should focus on validating these models in different populations and examining their integration with personalized intervention strategies to reduce dementia risk.

Aging; Dementia; Early diagnosis; Machine learning; Mil cognitive impairment; XGBoost; Dementia; Early diagnosis; Machine learning; Mil cognitive impairment; XGBoost