Rebar Price Prediction Using Stacking Methods: A Comparative Analysis of LSTM and NeuralProphet Fusion Models

Abstract

The fluctuation in rebar prices directly impacts the production costs in industries such as construction and metallurgy. Accurate price prediction is crucial for informed decision-making in these sectors. Traditional time series forecasting methods, however, are insufficient in capturing the nonlinear characteristics of price fluctuations, In recent years, deep learning-based methods, such as NP(NeuralProphet) and LSTM(Long Short-Term Memory Networks), have become popular in time series forecasting due to their strong sequence modeling capabilities. This paper proposes a stacking method to create a hybrid NP-LSTM forecasting model, combining NP and LSTM models to enhance the accuracy of rebar price predictions on a monthly basis. Compared to the individual NP and LSTM models, the NP-LSTM hybrid model combines the strengths of both, compensating for their weaknesses and providing more accurate predictions. This approach enhances both model performance and its stability and reliability in real-world applications, offering significant practical value. This study demonstrates the effectiveness of the stacking method for multi-model fusion, offering a novel approach to rebar price prediction.