A Personalized Music Intervention Framework for Elderly Mental Health Using SWPSI-KNN and Neural Collaborative Filtering Based on EEG Signals

Abstract

Under the trend of global aging, psychological problems such as depression and anxiety are becoming increasingly prevalent among the elderly population. Traditional intervention methods suffer from lagging emotional recognition and insufficient personalization. To improve the mental health problems of the elderly, this study innovatively combines the optimization of the temporal characteristics of EEG signals (power spectral density, time-frequency analysis, dynamic time regularization) with a collaborative recommendation mechanism. A music electrical signal data acquisition system for the psychological health of elderly people based on dynamic analysis of EEG signals has been developed. The system employs real-time EEG acquisition (1000Hz sampling rate), preprocessing (1-50Hz bandpass filtering, ICA-based noise removal), and feature extraction, utilizing an enhanced K-Nearest Neighbor (KNN) algorithm (with sliding windowing and dynamic weight adjustment) to predict EEG responses under music intervention. Experiments involved 80 elderly participants from a nursing home, with datasets including baseline anxiety scales and EEG recordings, validated through randomized controlled trials. The results indicated that the model reduced the EEG tracking error (MAE) from the traditional KNN of 3.24 μV to 0.07 μV. The NCF mechanism achieved 93.2% accuracy in anxiety state classification. In practical applications, the anxiety relief efficiency reached 96.21%, compared to 72.5% in the control group, and the user satisfaction score was 9.5/10. By dynamically optimizing temporal features through dynamic time warping and real-time EEG feedback-driven music adjustment, the system enables personalized intervention, offering an innovative solution combining real-time monitoring and precision adjustment for elderly mental health.