A Customized Deep Sleep Recommender System Using Hybrid Deep Learning.

Area of Science:

• Artificial Intelligence
• Health Informatics
• Personalized Medicine

Background:

• Sleep quality is crucial for modern well-being and work efficiency.
• Personalized sleep services require accurate sleep analysis, but individual patterns vary, and new users present a cold start problem.
• Existing recommendation systems lack standardization for sleep patterns and struggle with new user data.

Purpose of the Study:

• To develop a Customized Deep Sleep Recommender System (CDSRS) addressing individual sleep variations and the cold start problem.
• To enhance personalized sleep services through a hybrid learning approach.
• To improve the accuracy and user satisfaction of sleep recommendations.

Main Methods:

• Utilized K-means clustering for sleep pattern definition.
• Employed a hybrid learning algorithm combining user-based and collaborative filtering.
• Incorporated feedback top-N classification for user profile learning and recommendation refinement.
• Collected data from mobile devices and AI motion beds (snoring, sleep time, movement, noise).

Main Results:

• The hybrid learning approach demonstrated 13.2% lower Mean Squared Error (MSE) than collaborative filtering (CF) and 10.2% lower than content-based filtering (CBF).
• The system achieved 14.7% higher accuracy than CF and 9.2% higher accuracy than CBF based on Mean Absolute Percentage Error (MAPE).
• CDSRS showed superior performance in reflecting user evaluations and increasing recommendation accuracy with more users.

Conclusions:

• CDSRS, using a hybrid learning method, provides more accurate and customized sleep recommendations than traditional CF, CBF, and combination models.
• The system effectively addresses the cold start problem and adapts to individual sleep patterns.
• The proposed system enhances personalized sleep services, improving user well-being and efficiency.