Smart vertical farms have garnered significant attention for their ability to utilize urban space efficiently while enabling year-round, stable crop production. However, high energy consumption and elevated operational costs remain major challenges, making it essential to explore effective strategies for improving efficiency. This study integrates approximately nine months of data collected from a real-world vertical farm in Korea, combining environmental variables such as temperature, humidity, wind speed, and soil moisture with operational variables including electricity usage, labor input, water consumption, and production cost into a unified dataset. The dataset was refined through preprocessing steps including missing value and outlier removal, categorical encoding, scaling, and time-series segmentation. Three machine learning models Random Forest, XGBoost, and LSTM were then applied and compared. The results indicated that XGBoost consistently demonstrated the lowest prediction errors, showing strong performance in forecasting key indicators such as energy use and production costs. However, at certain points where sudden fluctuations occurred, LSTM exhibited more precise prediction capabilities due to its time-series specialization. Overall, this research highlights that a machine learning approach which simultaneously considers both environmental factors (e.g., temperature and humidity) and operational variables (e.g., energy, labor, and nutrient solutions) can offer a practical solution for enhancing energy efficiency and reducing costs in vertical farm operations.