Predictive analytics for inventory optimization

Abstract

In 2021, the Sri Lankan apparel manufacturing industry faced a severe downturn due to the COVID-19 pandemic and economic crises, highlighting the need for accurate sales predictions amid global supply chain disruptions. Traditional statistical models struggle to handle such crises, necessitating the exploration of machine learning methods for forecasting sales. This study aimed to identify the most effective predictive models for finished apparel goods sales, addressing data complexities like seasonality, trend, and stationarity, with a focus on enhancing decision-making in the industry. The dataset consisted of 128 weekly records of point-of-sale (POS) data for three specific apparel items sold in the US and manufactured in Sri Lanka, spanning from January 2021 to June 2023. Also, the inflation rate in the USA is used as an exogenous variable. Data preprocessing began with rationalization, followed by splitting it into training and testing sets. Two models, ARIMA and SARIMAX, were constructed using the training data to analyze the time series. Model performance was assessed using Mean Square Error (MSE), with the goal of generating future sales predictions. The results indicated that the ARIMA model outperformed SARIMAX, exhibiting significantly lower MSE values. This outcome suggests that ARIMA is the superior model for forecasting sales in this context. Future research aims to validate this result by incorporating additional datasets, ensuring the continued effectiveness of the ARIMA model in predicting apparel sales. In conclusion, this study highlights the critical role of advanced machine learning techniques, in improving sales predictions for the Sri Lankan apparel manufacturing industry. By addressing data complexities and employing robust validation methods, this research contributes to more precise planning and decision-making, essential for navigating disruptions in the global supply chain and economic uncertainties.