Scholarlink

<p>Predictive analysis plays a vital role in modern data-driven decision-making, where accurate models enable actionable insights across diverse domains, including healthcare, finance, and engineering. Poor configuration of parameters and computational inefficiencies constitute frequent problems in predictive analytics utilizing support vector machines (SVM). Existing hybrid models, such as particle swarm optimization integrate with support vector machine (PSO-SVM), differential evolution integrate with support vector machine (DE-SVM), and crow search algorithm combine with support vector machine (CSA-SVM), enhance performance but encounter difficulties such as premature convergence, a lack of exploration–exploitation balance, and vulnerability to local minimum. This study proposes an improved triple-hybrid framework of the PSO-DEA-CSA-SVM model for optimizing SVM hyperparameters, incorporating the strengths of particle swarm optimization (PSO), differential evolution algorithm (DEA), and crow search algorithm (CSA). By combining exploration, exploitation, and adaptive mutation, the model achieves faster convergence, greater robustness, and improved predicted accuracy across both benchmark and real-world datasets. The hybrid optimization approach enhances SVM’s decision boundary, enabling it to effectively separate classes despite noisy and imbalanced data. As a result, the model achieves higher classification accuracy, precision, sensitivity, F1-score, and lower classification error compared to standalone or traditional SVM models. This leads to faster convergence toward optimal solutions, reducing training time while maintaining high classification performance. To evaluate the proposed model, a liver cirrhosis dataset was utilized for testing and validation. The results demonstrated that the proposed model with three metaheuristics achieved better accuracy and outperformed the other six hybrid models. The PSO-DEA-CSA-SVM model achieved the highest score of accuracy of 99%, precision 99%, sensitivity 98%, F1-score 98%, and ROC-AUC of 100% in the liver cirrhosis model. The results indicate that the proposed method is a promising tool for predictive analysis, offering improved classification, accuracy, and reliability. In contrast, the proposed method outperformed some other similar existing methods.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An enhanced hybrid framework for predictive analytics utilizing metaheuristic optimization techniques

  • 期刊论文

  • 发表日期：

    2026年2月14日

查看全文
  • Samaila Abdullahi,
  • Saratha Sathasivam

摘要

Predictive analysis plays a vital role in modern data-driven decision-making, where accurate models enable actionable insights across diverse domains, including healthcare, finance, and engineering. Poor configuration of parameters and computational inefficiencies constitute frequent problems in predictive analytics utilizing support vector machines (SVM). Existing hybrid models, such as particle swarm optimization integrate with support vector machine (PSO-SVM), differential evolution integrate with support vector machine (DE-SVM), and crow search algorithm combine with support vector machine (CSA-SVM), enhance performance but encounter difficulties such as premature convergence, a lack of exploration–exploitation balance, and vulnerability to local minimum. This study proposes an improved triple-hybrid framework of the PSO-DEA-CSA-SVM model for optimizing SVM hyperparameters, incorporating the strengths of particle swarm optimization (PSO), differential evolution algorithm (DEA), and crow search algorithm (CSA). By combining exploration, exploitation, and adaptive mutation, the model achieves faster convergence, greater robustness, and improved predicted accuracy across both benchmark and real-world datasets. The hybrid optimization approach enhances SVM’s decision boundary, enabling it to effectively separate classes despite noisy and imbalanced data. As a result, the model achieves higher classification accuracy, precision, sensitivity, F1-score, and lower classification error compared to standalone or traditional SVM models. This leads to faster convergence toward optimal solutions, reducing training time while maintaining high classification performance. To evaluate the proposed model, a liver cirrhosis dataset was utilized for testing and validation. The results demonstrated that the proposed model with three metaheuristics achieved better accuracy and outperformed the other six hybrid models. The PSO-DEA-CSA-SVM model achieved the highest score of accuracy of 99%, precision 99%, sensitivity 98%, F1-score 98%, and ROC-AUC of 100% in the liver cirrhosis model. The results indicate that the proposed method is a promising tool for predictive analysis, offering improved classification, accuracy, and reliability. In contrast, the proposed method outperformed some other similar existing methods.