Testing the auto-regressive integrated moving average approach vs the support vector machines-based model for materials forecasting to reduce inventory

Abstract

Poor planning and scheduling increase buying, storage, and obsolescence expenses. Material shortages increase labor, machine optimum time, etc. Industrial raw materials, semi-finished items, spares, and consumables have distinct consumption patterns, reorder points, purchase lead times, quantity limits, discounts, etc. To save money, machine learning predicts demand and prepares materials. This study employs ARIMA or Support Vector Machine (SVM) machine learning-based forecasting approaches to forecast materials for less inventory. Feature engineering eliminates seasonality, time series, and external demand and ignores data irregularities, missing figures, and disparities. This approach needs to adapt traits to factors, separate test and training data, and consider many future models to represent the best forecasts. Forecast reliability and consistency were examined for each model. Inventory management systems were evaluated for computational complexity and installation ease and found implementation issues. Both models’ input data and resilience were examined using sensitivity analysis. Accurate prediction SVM and ARIMA predict material demand differently. Meaningful statistics show the optimal model. Performance differences between SVM and ARIMA enhance model selection. Thinking about the execution of high inventory system integration and computational complexity, response surface methodology chooses factorial variables with the highest or lowest responses. Analysis of variance, factor analysis, and effect modeling expansions demonstrated for the response.