a novel approach for multi product demand forecast using data mining techniques (empirical study: carpet industry)

Accurate demand forecasting plays an important role in meeting customers’ expectations and satisfaction that strengthen the enterprise's competitive position. in this research, time series and artificial neural networks methods compete to provide more precise demand estimation while having a large variety of products. after obtaining the initial results, suggestions have been implemented to improve forecasting accuracy. as a direct result of that, the average mean absolute percentage error (mape) of all products' demand forecast reduces significantly. to improve the quality of historical records, association rules and substitution ratio have been applied. this method plays a significant role to detect the existing pattern in historical data and mape reduction. the satisfactory and applicable results provide the company with a more accurate forecast. moreover, the issue of precepting confusing historical data which caused unforecastable trends has been solved. the r language and “neuralnet”, “nnfor”, “forecast”, and “arules” packages have been applied in programming.