Effective demand forecasting and inventory optimization are crucial to enhance supply chain effectiveness, reduce cost, and eliminate stock imbalance. This research investigates the application of deep learning algorithms Long Short-Term Memory (LSTM), Gated Recurrent Units (GRU), Temporal Convolutional Networks (TCN), and Temporal Fusion Transformers (TFT) to predict demand and optimize inventory for automobile tires. The information, collected from three well-known tire producers, distributors, and suppliers, over a specific time period, was preprocessed, 70% for training and 30% for testing. Each model was evaluated based on precision, recall, F1-score, accuracy, Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and forecasting accuracy. Results showed that the highest performing network was TFT with the lowest MAE (2.9) and RMSE (4.7) and the highest accuracy of 96.1%. TCN followed closely at 93.5%, while LSTM and GRU were lower relatively at 91.2% and 89.7%, respectively. The confusion matrix also favored TFT as the best of all the models, with 910 positive true predictions and only 40 false negatives, minimizing errors in forecasting and maximizing inventory levels. The findings highlight the potential of deep learning models, especially TFT, in improving inventory control and forecasting demand in supply chain management. Employing such advanced models can enable improved decision-making, reduced operational costs, and lessened supply chain disruption.