Application of K-Nearest Neighbor Regression Methods in Predicting Data Sales Suhendra and Siti Aisyah
Politeknik Negeri Media Kreatif, Indonesia
Abstract
CV. XYZ is a company specializing in the retail sales of women^s clothing. One of the challenges faced by the company is the difficulty in predicting which products will be popular among customers. To address this issue, it is essential to use the k-nearest neighbor regression algorithm to forecast product sales. The goal of this prediction is to assist the company in identifying products that are highly favored by customers by analyzing sales data and predicting future sales trends at CV. XYZ. Additionally, this research aims to evaluate the prediction model^s accuracy by calculating the error rate. The method employed in this study is the k-nearest neighbor regression, and the data processing stages are conducted using the Knowledge Discovery in Database (KDD) methodology. The results of the sales predictions for women^s clothing products at CV. XYZ using the k-nearest neighbor regression method with different K values are as K = 3 produced an RMSE value of 0.22607, with predictions for the Jumpsuit product in April. K = 6 produced an RMSE value of 0.27569, with predictions for the Homeware product in February. K = 8 produced an RMSE value of 0.29608, with predictions for the Muslim wear product in March. K = 2 produced an RMSE value of 0.34004, with predictions for the Dress product in April. K = 4 produced an RMSE value of 0.35204, with predictions for the Skirt product in September. K = 11 produced an RMSE value of 0.37841, with predictions for the Blouse product in November. K = 11 produced an RMSE value of 0.39944, with predictions for the Scarf product from January to December. K = 9 produced an RMSE value of 0.39598, with predictions for the Coat product in December. K = 14 produced an RMSE value of 0.40819, with predictions for the Tunic product in December. K = 14 produced an RMSE value of 0.47897, with predictions for the Suit product in August and December. K = 12 produced an RMSE value of 0.50079, with predictions for the Blazer product in July. K = 14 produced an RMSE value of 0.51028, with predictions for the Pants product in November. It is concluded that the average error rate across all models falls within a low error range, specifically between 0.00 - 0.299.
