A hybrid time series forecasting model combining recurrent neural networks and ensemble learning for furniture sales prediction

Abstract

This study proposes a deep learning model, named MP-LRNet, that aims to improve the accuracy and stability of time series forecasting by combining recurrent neural networks and ensemble learning algorithms within a modular Multi-Patch structure. The model is designed to learn temporal patterns at different time intervals and to capture both short- and long-period dependencies in sequential data. Accurate time series forecasting plays a central role in supporting decisions across various practical domains such as retail, production, and energy management. To evaluate the performance of MP-LRNet, experiments were conducted using a real furniture sales dataset and a publicly available energy consumption benchmark (UCI Household Electric Power Consumption). The proposed model achieved an 𝑅2 value of 0.9918, demonstrating reliable predictive ability and consistent results across different configurations. The Multi Patch structure enhanced temporal representation, while integrating long short-term memory and Random Forests improved predictive precision without a significant increase in computational time. The findings indicate that MP-LRNet serves as an effective approach for sales prediction and energy demand estimation, suggesting strong potential to be adapted for broader diverse applications, such as environmental analysis, in future studies