Assessment of atherosclerosis risk in an insufficient sample size based on K-Means BS and TW-gcForest.

Abstract

Background: Insufficient data is a common issue encountered in studies of atherosclerosis risk assessment. However, when the sample size is insufficient, commonly used classification algorithms often fail to achieve superior performance, thereby limiting the application of atherosclerosis data in patient risk assessment.

Purpose: In cases where the sample size is inadequate, the use of an algorithmic model can allow for an effective evaluation of the risk of atherosclerosis in patients.

Methods: We propose an oversampling technique called K-Means-Borderline-SMOTE (K-Means BS) and a classification algorithm named triple-weighted gcForest (TW-gcForest). Our proposed K-Means BS generates diverse synthetic samples by imposing strict constraints and avoids generating similar synthetic samples. TW-gcForest is mainly designed to address the problem of unfair forest weight allocation and sliding window weight allocation in standard gcForest. We perform numerical simulations on two datasets to demonstrate the robustness of the two methods.

Results: Numerical simulations show that standard gcForest achieves the high-performance for atherosclerosis risk assessment on the K-Means BS synthetic dataset. However, TW-gcForest exhibits superior performance to the standard gcForest on the original dataset, as well as the SMOTE and K-Means BS synthetic datasets.

Conclusion: Our approach can effectively improve accuracy, precision, recall, F1 score, and AUC compared with traditional algorithms.