Instructions for Preparing Papers for Proxies:Jurnal Informatika

Abstract

This study aims to compare the performance of linear and nonlinear regression algorithms in predicting monthly store revenue for the next two years to support business decision making and prevent bankruptcy risk. The models used include Linear Regression, Quadratic Non-Linear Regression, and Support Vector Regression (SVR) with Polynomial, RBF, Linear, and Sigmoid kernels, with evaluation using MSE, RMSE, and percentage as evaluation metrics. The results show that Linear Regression provides the best performance with the lowest error value, where the error percentage is in the range of 1.30% to 3.13% in five stores (S001–S005), making it effective in predicting revenue for the next two years in this case and dataset. The Quadratic Non-Linear model produces poor predictions, even producing negative values due to its mathematical nature, while Linear SVR and RBF show quite good performance although RBF tends to produce flat predictions. Polynomial SVR provides less than optimal results due to the characteristics of the data which is linear or close to linear. Overall, Linear Regression proved to be the most effective in this case, with the research limitations being that the data period was only two years and that time series methods had not been applied to capture seasonal patterns or specific events.