Prediction of yarn sales price using data mining techniques – a case of yarn manufacturing industry

Abstract

Data-driven knowledge is required for businesses to make better decisions that result in profit maximisation. In this study, it has been attempted to develop a model to predict yarn sales prices against cotton prices and other parameters. For this purpose, four different data mining techniques namely ARIMA (Autoregressive Integrated Moving Average), Multivariate regression, K-Nearest Neighbor (KNN) and Neural Networks (NN), were considered. The entire analysis was performed on thirty months of data that was collected from the ERP system of a yarn manufacturing industry. The unique aspect of this study is that before separately deploying data mining techniques, significant parameters that impact yarn sales prices were identified through Adjusted R-squared values. Seasonal and trend patterns were checked on yarn sales data, and seasonal adjustments were obtained through data mining algorithms. The performance of all four models was evaluated using Mean Absolute Error and Root Mean Square Error. The analysis shows that the KNN model, in the stated settings, is the most accurate as evident from MAE and RMSE values of 222.85 and 285.082, respectively. This study’s unique combination of features and machine learning algorithms is envisaged to be valuable for decision-makers in the textile yarn manufacturing industry.