Proyeksi Tren Kategori Pakaian Mendatang Menggunakan Random Forest pada Data Transaksi Pelanggan

Abstract

The dynamic fashion industry requires accurate trend projections for marketing and product development. This study aims to project future clothing trends using Random Forest as the primary model and XGBoost as the secondary model. The main dataset contains 3,900 transactions with demographic information, purchase history, seasonal data, and product categories. For local context, inventory data from the “Coffer Ruh” fashion store was integrated as a companion case study. The methodology included preprocessing, handling class imbalance with SMOTE, stratified splitting (80:20), training Random Forest and XGBoost, and evaluation using accuracy, precision, recall, F1-score, and a confusion matrix. Evaluation results (Outerwear, Footwear, Bottoms, Tops, Accessories) show that Random Forest achieved an accuracy of 67.56%, weighted precision of 68.04%, recall of 67.56%, and an F1-score of 67.79%, while XGBoost demonstrated similar performance with an accuracy of approximately 68%. The Random Forest model projected Jackets (17%), Coats (12%), and Shoes (10%) as the top three global trend categories. Store data analysis revealed the highest stock levels for children’s masks (35 pcs), red cornersticks (33 pcs, coats), and drams (31 pcs, jackets). There is some alignment: two of the three products with the highest inventory are outerwear items that align with global trends; however, masks are not a predicted apparel category. Due to limitations in the store data (small sample size, lack of time/transaction dimensions), transfer learning or hybrid dataset approaches cannot yet be applied, which is identified as a limitation and a direction for future research.