Performance Analysis And Prediction Model Evaluation For Dost-Assisted MSMEs Using Machine Learning

Predicting business performance is a significant area of research in both business and economics, as successful businesses have a big impact on the overall economic development. This research has developed a business performance prediction model that uses data mining techniques to classify applicants to the DOST Small Enterprise Technology Upgrading Program (SETUP) based on their likelihood of graduation. This study investigates factors that affect business performance using historical data from 503 previous SETUP adopters and runs different predictive models, such as Naïve Bayes, Random Trees, Support Vector Machines (SVMs) and Decision Trees, which are tested for training accuracy. Main findings revealed significant factors in SETUP-assisted MSMEs' performance are the company's location, sector industry, business classification and total sales over total liabilities ratio. With a prediction accuracy of 79.60%, the J48 model with CFSSubset Feature Selection yielded the best results using the data of SETUP applicants in the identified factors. This method was pilot tested during the SETUP applicant evaluation, which led to a better decision for the implementers in selecting MSMEs that will successfully implement the program.