An Impact Investigation of Data Normalization Methods on Load Forecasting with ANN Model

Data normalization is a crucial pre-processing step in machine learning, where features are scaled or transformed to ensure that each contributes equally to the model. The effectiveness of machine learning algorithms mostly relies on the quality of the data, which directly impacts the model's ability in forecasting tasks. While numerous studies have highlighted the effects of normalization in enhancing data quality and algorithm performance, there has been limited focus on the combined impact of feature selection and normalization-a current research area aimed at boosting model accuracy. This study specifically explores how different normalization techniques affect forecasting performance, using electrical and meteorological datasets of Chhattisgarh state, covering the period from January 2022 to January 2024. The investigation involves applying these normalization methods and assessing their impact on forecasting accuracy using an Artificial Neural Network (ANN) model. Among the techniques evaluated, z-score normalization consistently produces superior results compared to Minimum-Maximum Normalization and Maximum Normalization methods. The findings were further validated by calculating error metrics such as Mean Absolute Percentage Error (MAPE) and Mean Absolute Error (MAE) against the actual load for each normalization method. Specifically, the z-score normalization method achieved the lowest error, with MAPE of 3.449% and MAE of 150.622 MW for the month of January 2024.