A Sporadic Classification and Regression-Based Approach to Intermittent Demand Forecasting in Smart Supply Chain

Intermittent demand forecasting presents a distinct problem in supply chain management, as it requires accurate prediction of demand in order to minimize costs and enhance operational efficiency for businesses. The present study introduces a novel data-driven approach for handling the problem of forecasting intermittent demand combinations across several time horizons. The approach involves building an efficient sporadic classification model and using regression techniques to predict the quantity of non-zero demand for future time horizons. This unique two-stage forecasting framework is based on the implementation of the best threshold classification methods using LGBM. The results show a significant improvement in classification accuracy for splitting intermittent requests. The output from the first phase has been given to the multimodal temporal attention-based Seq2seq approach, which prioritizes various aspects of the past in order to predict multiple future time series over different time horizons. The experiment results were obtained using the Corporación Favorita dataset, which was made publicly available for a Kaggle competition. Our approach has demonstrated good performance when compared to state-of-the-art techniques. The findings demonstrate that this study can also offer precise Smart inventory relies on upstream inputs to ensure accurate and efficient decision-making in the smart supply chain.