AlF3 feeding is crucial for maintaining pot thermal balance and electrolyteElectrolyte temperature. The current reactive approach lacks precisionPrecision, often causing additional feeding and in some cases insufficient feeding due to certain factors like misattributed temperature rises, errors in bath chemistry and operational inefficiencies. This improper thermal control negatively impacts pot performance. This article presents a predictive modelPredictive model for optimizing AlF3 shots in aluminum smeltingAluminum smelting operations, leveraging Machine LearningMachine learning techniques specifically through Artificial Neural Network (ANN) & Support Vector Regression (SVR). Using data from a 360kA pot-line, the model predicts AlF3 shots at day level for 360 pots across different pot-age groups—low, intermediate, mid, and old—ensuring precise and tailored AlF3 feeding strategies. The SVR and ANN based model forecasts AlF3 requirements by analyzing key performance indicators and operational metrics, focusing on minimizing Mean Absolute Percentage ErrorMAPE (MAPE). This predictive modelPredictive model aids in maintaining pot temperature and controlling excess AlF3 with minimal deviation, thereby enhancing stability and improving current efficiencyCurrent efficiency.