ScholarWorks@성균관대학교

초록

This study presents a lifecycle-based clustering approach to analyze weekly sales data, aiming to enhance sales prediction accuracy for small-scale retailers. By combining time series clustering and deep learning methods, the proposed framework classifies sales patterns into four distinct lifecycle stages and predicts future cluster memberships. Using franchise store card and delivery sales data from an anonymous startup's app, we demonstrate that integrating Self-Organizing Maps (SOM) with TimeSeriesKMeans (TSKMeans) improves clustering accuracy compared to using TSKMeans alone. Additionally, the Temporal Transformer Model (TTM) is applied to predict cluster classifications, achieving a validation accuracy of up to 89% over a 24-week period. The results underscore the effectiveness of lifecycle-based clustering for enhancing sales forecasting in dynamic and uncertain market environments, offering valuable insights for small retailers looking to optimize their sales strategies.

키워드