COMPARATIVE NEURAL NETWORK MODELS FOR CUSTOMER DEMAND PREDICTION: ENHANCING INVENTORY MANAGEMENT IN ADDIS MACHINE SPARE PART AND MANUFACTURING INDUSTRY.

Abstract

Effective inventory management is essential for businesses in today’s competitive global market. By accurately forecasting demand, companies can determine optimal stock levels, enhancing both efficiency and profitability. However, the Addis Machine Spare Part and Manufacturing Industries (AMSPMI’s) outdated manual inventory management system struggles to keep up with market demands, leading to frequent overstock and understock of products, disrupting production and customer satisfaction. This challenge hinders the company’s ability to compete globally. To address these logistical challenges, this paper offers a comparative analysis of four leading time series demand forecasting models aimed at improving inventory management: Artificial Neural Network (ANN), Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU). The models' accuracy was evaluated using Mean Squared Error and Root Mean Squared Error. Based on the results of the models, the LSTM model outperformed with the highest prediction accuracy of 99.95 % with the least square error of 0.05 %. According to the analysis, using the recommended strategy would be advantageous since it would significantly reduce the issues related to excess stock and stock outs. By managing both surplus stock and stock-outs, businesses may achieve a more balanced and efficient inventory management and improve their overall financial performance, operational efficiency, and customer satisfaction. Therefore, the study highly recommends that this manufacturing company manage its inventory using the LSTM model