An Adaptive Forecasting Framework for Time-Varying Sales Data Using Multi-Resolution Learning Models

Abstract

Time-varying sales data, characterized by non-stationarity and dynamic trends, presents a significant challenge for accurate forecasting. This paper proposes an adaptive forecasting framework that leverages multi-resolution learning models, integrating hierarchical temporal features to enhance prediction accuracy. The framework employs hybrid machine learning techniques—combining temporal decomposition with deep learning regressors—to adaptively capture structural shifts in sales patterns. Experimental results on real-world retail datasets demonstrate the superiority of the proposed model over traditional forecasting approaches, particularly under conditions of volatility and seasonality.