DOI: https://doi.org/10.36347/sjet.2025.v13i10.007

Pages: 845-854

AI-based forecasting models have revolutionized industries, enabling efficient operations and enhanced sustainability. However, understanding their comparative performance across multiple domains remains underexplored. This study evaluates AI-based forecasting models' performance in smart manufacturing, inventory planning, and sustainability trends, focusing on accuracy, stability, and domain-specific applicability. Conducted at Lamar University, USA, this research spanned from January 2023 to June 2024. The study sample consisted of 42 data sets derived from smart manufacturing systems, inventory usage, and sustainability trends. Machine learning algorithms, including Random Forest, LSTM, and Support Vector Machines (SVM), were applied for forecasting power consumption, raw material demand, and sustainable behavior shifts. The models were evaluated using accuracy, standard deviation, p-values, Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Time Series Performance Metrics. The results highlighted that the Hybrid LSTM model for smart manufacturing (power consumption) achieved the highest accuracy of 97.6%, a low standard deviation of 0.03, and a p-value of 0.002, indicating statistical significance. The MAE and RMSE for this model were 0.15 and 0.22, respectively. In inventory planning, the Random Forest model provided a robust forecast with an accuracy of 92.4%, standard deviation of 0.05, and a p-value of 0.01. The model's MAE was 0.12 and RMSE 0.18, demonstrating its reliability. For sustainability trend forecasting using social signals, the SVM model achieved an accuracy of 85.8%, a standard deviation of 0.07, and a p-value of 0.04. The MAE was 0.25, and RMSE 0.30. The analysis revealed that, while LSTM models performed best for time-series and continuous data (manufacturing), Random Forest models excelled in discrete demand forecasting, and SVM models were more suited for signal-based, nontraditional data forecasting.