Factors affecting subjective cognitive decline: an automated machine learning approach

This study aims to develop a screening model for subjective cognitive decline (SCD) based on machine learning techniques.

A retrospective cohort study collected clinical psychological factor data from the "Active Health" screening app under the National Key R&D Program. The final dataset included 598 samples, with an SCD incidence rate of 26.12%. The data were randomly divided into a training set (n = 418). A validation set (n = 180) at a ratio of 7:3. In the training set, prediction models for SCD were constructed using logistic regression (LR), Naive Bayes, support vector machine (SVM), decision tree, and neural network algorithms. Model performance on the validation set was assessed by calculating the area under the ROC curve (AUC), accuracy, sensitivity, specificity, precision, recall, and F1 score. SHAP values were used for model interpretability analysis.

The SVM model showed good performance in the training set, with an AUC of 0.82, indicating strong predictive ability. Information Overload (IO), Self-Perception (SP), Energy Level (EL), Depressive Emotion (DE), Gender (SEX), Risk Decision (RD), and Short-Term Memory (STM) were important feature variables for SCD occurrence.

This study successfully developed an SVM-based model for screening the risk of SCD. The SVM model demonstrated superior predictive performance compared to Naïve Bayes, Decision Tree, Neural Network, and traditional LR models.