Comparison between several machine learning algorithms to evaluate socioeconomic influences on hydroelectric generation in Malaysia

Accurate forecasting of hydroelectric power generation is important for supporting long-term energy planning and improving the management of renewable energy resources in Malaysia. While many previous hydropower forecasting studies emphasise hydrological variables such as rainfall and reservoir inflow, the potential influence of socioeconomic indicators on hydroelectric generation has received comparatively less attention. However, hydroelectric forecasting remains challenging due to complex and nonlinear relationships between influencing factors, which are often not fully captured by traditional methods. In addition, limited studies have compared machine learning models for hydroelectric forecasting in the Malaysian context, particularly when incorporating socioeconomic indicators. This study aims to analyse the relationship between selected socioeconomic indicators and hydroelectric power generation in Malaysia and to evaluate the performance of several machine learning models for forecasting hydropower output. Annual data covering the period from 1980 to 2021 were collected for Gross Domestic Product (GDP), energy consumption, population, inflation rate, and hydroelectric generation. Three Artificial Neural Network (ANN) configurations were first developed using different input combinations: Model A (GDP and energy consumption), Model B (GDP, energy consumption, and population), and Model C (GDP, energy consumption, and inflation). To assess the reliability of the input configurations, a five-fold cross-validation procedure was applied. The results showed that Model B produced the most favourable cross-validation performance, achieving an average Mean Squared Error (MSE) of 1.8955 × 10⁵ with an R-value of 0.7731. Based on this result, the selected input combination was further evaluated using four machine learning models, namely Artificial Neural Network (ANN), Support Vector Machine (SVM), Random Forest, and XGBoost. Among the evaluated models, ANN demonstrated the most consistent predictive performance, achieving a testing MSE of 1.1541 × 10⁴ and an R-value of 0.9962. These results suggest that socioeconomic indicators such as GDP, energy consumption, and population may provide useful explanatory signals for hydroelectric generation forecasting. However, the findings should be interpreted cautiously due to the relatively limited dataset used in this study. Despite this limitation, the study demonstrates the potential of machine learning approaches for supporting data-driven energy forecasting and planning.