Demand forecasting using KAN-RNN

Abstract

In today’s highly competitive business environment, organizations continuously strive to maintain their competitiveness and achieve sustainable profit margins to support long-term growth and development. Accurate demand forecasting has become a critical tool for decision-makers, as it allows better resource allocation, inventory management, and strategic planning. Recurrent deep learning methods, which use gating mechanisms to maintain an internal state aligned with time series data, are among the most widely used approaches to improve forecast accuracy. Despite their success, these models still exhibit significant untapped potential that could be realized by rethinking the design of their gating mechanisms. To address this, we introduce a novel demand forecasting method inspired by Kolmogorov–Arnold networks (KANs), featuring a modified recurrent architecture with a restructured gating mechanism. This innovation leverages KAN principles to enhance the model’s capacity to capture intricate temporal dependencies and adapt to evolving demand patterns. Experimental evaluations demonstrate that the proposed method outperforms state-of-the-art approaches, highlighting its ability to provide more accurate and reliable demand forecasting results.