Comparative Analysis of Artificial Neural Networks and Multiple Linear Regression in Electrical Load Forecasting under Limited Data Conditions: A Case Study at Cu Chi Power Company

Medium-term electrical load forecasting plays a pivotal role in planning investment, upgrading, and operating distribution grids in urbanizing areas. This study aims to evaluate and quantify the effectiveness of traditional statistical mathematical models versus modern machine learning algorithms in the context of limited planning data at the local level. Based on a 16-year historical dataset (2010 - 2025) in Cu Chi district, three methods including multiple linear regression (OLS), time series (ARIMA), and artificial neural networks (ANN) were es Table III lished to interpolate and extrapolate forecasting scenarios for the 2026-2030 period. Experimental results indicate that, with the characteristic of a small sample size (N=16), the machine learning algorithm suffers from severe overfitting, causing the future forecasting trajectory to be illogically saturated, despite a very low training error of 1.24%. In contrast, the multiple linear regression method maintains high reliability with a mean absolute percentage error of 4.12%, and the future interpolation results are completely consistent with the macroeconomic growth momentum. Consequently, the multiple linear regression method is proven to be the most explicit and reliable tool for load forecasting, directly supporting sustainable energy infrastructure investment decisions under limited data conditions.