Novel AI-Powered Dynamic Inventory Management Algorithm in the USA: Machine Learning Dimension

Abstract

Dynamic inventory management revolves around the practice of progressively modifying inventory degrees to adapt to fluctuations in client demand, production, and supply chain dynamics. At the center, inventory management focuses on upholding enhanced levels of stock to balance consumer service via availability with the costs related to holding excess inventory. This research paper aimed to explore the dynamic inventory management activities employed by organizations in the USA, shedding light on the machine learning strategies that can be deployed and their implications. The performance of the algorithms was empirically evaluated in a Python program experiment utilizing real-world data. To facilitate the data for input into the Neural Network, feature engineering, and selection were imposed to affirm its suitability. This study proposes the Sequence-to-Sequence (Seq2Quant) algorithm, a neural network-powered technique for demand prediction in inventory management. The current experiment compared and contrasted the performance of the Neural Networks against the following baselines, most notably, Naïve Seasonal Forecast, Moving Average Forecast, ARIMA, Naïve Seasonal Forecast with Averaging over four periods, SARIMAX. From the experiment, it was evident that the Seq2Seq had the lowest MAE (17.44) and the lowest SMAPE (66.91), suggesting that it was the best-performing algorithm overall. Besides, SARIMAX and ARIMAX also performed well, with MAE values of 18.33 and 18.09, respectively.