An Intelligent Decision Support Model Using Hybrid K-Means PSO-SVM for MSME Development Priority Classification

Abstract

Addressing the issue of subjective bias inherent in conventional analytical data collection methods. Intelligent computing approaches are essential to ensure the precise allocation of local government assistance programs, such as Dumisake. This study aims to design and evaluate the performance of a hybrid K-Means PSO-SVM architecture for classifying MSME assistance priorities based on empirical financial and operational characteristics. Feature extraction for the model incorporates variables such as monthly turnover, production capacity, number of employees, and business tenure. The K-Means algorithm is employed in the initial stage to categorize data into three priority clusters: high, medium, and low. Subsequently, the Particle Swarm Optimization (PSO) algorithm is specifically implemented to automate the search for optimal parameters for the Support Vector Machine (SVM) classifier. Optimization results indicate that PSO iterations successfully identified a penalty parameter (C) of 46.1258 and a kernel width (γ) of 4.7327. This configuration significantly boosted the SVM model's performance, achieving a final accuracy of 98.04%, an F1 Score consistently above 0.97, and a negligible classification error rate. In conclusion, this hybrid prediction architecture demonstrates high reliability in clearly defining class boundaries, offering an objective decision support solution for local governments to select MSME aid recipients transparently and accurately.