Swarm Intelligence-Based Feature Selection and ANFIS Model Parameter Optimization for ASCV Risk Prediction and Classification

摘要

In the absence of Indian ethnic-specific cardiovascular (CV) risk prediction tools, machine learning models with artificial intelligence (AI) techniques are beneficial. This study focuses on the comparison of two intelligent CV risk prediction and classification models. The study has used both traditional and non-traditional CV risk markers to identify the Atherosclerotic cardiovascular (ASCV) risk status at an early stage. To handle the missing data, we have used multiple imputation (MI) using the gaussian copula (GC) method. This work has studied two popular swarm intelligence (SI) techniques for optimal feature subset selection and tuning the neuro-fuzzy learning process for ASCV risk prediction. In the proposed model, selection of optimal input feature and ASCV risk prediction is implemented using the Particle Swarm Optimization (PSO) and Grey Wolf Optimization (GWO) algorithms. Optimal feature selection was done using the fitness function-based evaluation of wrapper-based multi-support vector machine (multi-SVM) classifier. Secondly, the optimal features are fed to the adaptive neuro-fuzzy inference system (ANFIS), whose parameters are optimized using PSO and GWO, denoted as ANFIS_(PSO) and ANFIS_(GWO) for ASCV risk prediction. Finally, the risk predicted by SVM_(PSO)-ANFIS_(PSO) and SVM_(GWO)-ANFIS_(GWO) models are classified using a multi-SVM classifier and compared to identify the emerging robust model. The proposed framework is validated in MATLAB using Kerala-based clinical data. The final model performance of SVM_(PSO)-ANFIS_(PSO)-Multi-SVM has shown 88.41 % (training) and 95% (testing) accuracy, 79.02% (training), 89.47% (testing) sensitivity, and with 91.84% (training), 97.47% (testing) specificity the PSO model outperforms the SVM_(GWO)-ANFIS_(GWO)-Multi-SVM model showing higher performance variables.