An Ensemble Multilabel Classification for Disease Risk Prediction.

Area of Science:

• Medical Informatics
• Machine Learning
• Data Science

Background:

• Early identification and prevention of disease risk are essential for public health.
• Regular physical examinations provide valuable data for disease risk assessment.
• Multilabel classification presents challenges, particularly with imbalanced datasets.

Purpose of the Study:

• To propose a novel method for disease risk prediction formulated as a multilabel classification problem.
• To address the imbalance learning problem in disease risk prediction.
• To evaluate the performance of the proposed method against existing techniques.

Main Methods:

• A novel Ensemble Label Power-set Pruned datasets Joint Decomposition (ELPPJD) method was developed.
• The multilabel classification problem was transformed into a multiclass classification problem.
• Pruned datasets and joint decomposition methods, including size balanced (SB) and label similarity (LS) strategies, were employed to handle data imbalance.

Main Results:

• The ELPPJD method demonstrated superior performance compared to classic multilabel classification methods RAkEL and HOMER.
• The label similarity (LS) strategy within the ELPPJD method yielded outstanding results.
• Experimental validation was conducted using real-world physical examination records.

Conclusions:

• The proposed ELPPJD method is effective for disease risk prediction.
• The label similarity strategy is particularly beneficial for improving prediction accuracy in imbalanced datasets.
• This approach offers a promising direction for leveraging medical data for early disease detection.