Enhancing Machine Learning Models for Vertical Farm Energy Forecasting: Impact of Data Smoothing and Feature Selection

The widespread adoption of vertical farming is constrained by excessive energy consumption, highlighting the need for accurate energy consumption forecasting to develop effective energy-saving strategies. Data-driven models have become increasingly important for this purpose, yet prediction accuracy depends heavily on both data smoothing and feature selection. However, their effects on vertical farm energy estimation remain underexplored. This study examines how different data preprocessing methods and feature selection techniques influence energy cost prediction in vertical farming using data-driven models. Specifically, it compares two data smoothing techniques—Gaussian Kernel Density Estimation and the Savitzky-Golay filter—in the preprocessing stage. Additionally, it evaluates three feature selection methods: backward elimination, PCA-based backward elimination, and genetic algorithms (GA), assessing their impact on model performance. The results indicate that the Savitzky-Golay filter and PCA-based backward elimination significantly enhance both prediction accuracy and computational efficiency. These findings provide valuable insights for optimizing energy efficiency in vertical farming.