Predicting dyslipidemia incidence: unleashing machine learning algorithms on Lifestyle Promotion Project data

Area of Science:

• Biomedical Informatics
• Public Health
• Computational Biology

Background:

• Dyslipidemia, defined by abnormal plasma lipid profiles, is a significant global health concern responsible for millions of deaths annually.
• Early identification and intervention are critical for managing dyslipidemia and mitigating its widespread health impact.

Purpose of the Study:

• To predict the incidence of dyslipidemia using advanced machine learning techniques.
• To identify key predictive features and assess the performance of various machine learning models for dyslipidemia prediction.

Main Methods:

• Utilized a comprehensive dataset from the Lifestyle Promotion Project (LPP) in Iran, involving data preprocessing, merging, and null value handling.
• Applied normalization techniques and three feature selection algorithms to optimize predictive modeling for five dyslipidemia-related variables.
• Evaluated multiple machine learning algorithms, including Multi-Layer Perceptron neural network (MLP) and Random Forest.

Main Results:

• Multi-Layer Perceptron neural network (MLP) demonstrated superior performance in accuracy, F1 score, sensitivity, and specificity.
• Random Forest also achieved high accuracy and outperformed K-Nearest Neighbors (KNN) in precision, recall, and F1 score.
• Feature selection identified significant predictors of dyslipidemia, including waist circumference, serum vitamin D, blood pressure, sex, age, diabetes, and physical activity.

Conclusions:

• Machine learning models, particularly MLP, show significant potential for accurate dyslipidemia incidence prediction.
• The study highlights the complex interplay of various factors contributing to dyslipidemia, reinforcing the value of machine learning in understanding and predicting its occurrence.
• Identified key risk factors provide insights for targeted public health interventions and personalized medicine approaches.