EEMD-LASSO-LSTM: Hybrid Model for Time Series Forecasting
📌 Official Code for the Paper: "Methodology for Advanced Time Series Demand Forecasting: A Hybrid Model of Decomposition and Deep Learning"
📌 Authors: Juyoung Ha, Sungwon Lee, Sooyeon Jeong, Doohee Chung
🚀 Overview
This repository contains the implementation of the EEMD-LASSO-LSTM model, a novel hybrid approach for demand forecasting. Our methodology integrates:
- EEMD: Decomposes time-series data into multiple frequency components.
- LASSO: Selects the most relevant features.
- LSTM: Captures long-term dependencies in demand fluctuations.
Model Architecture
Below is the architecture of our proposed model:
📊 EEMD Decomposition Results
Our model first applies Ensemble Empirical Mode Decomposition (EEMD) to break down the original time-series data into multiple Intrinsic Mode Functions (IMFs) and a residual component.
📈 Benchmark Results
The following table summarizes the performance of EEMD-LASSO-LSTM compared to various baseline models:
✅ EEMD-LASSO-LSTM achieves the lowest NRMSE across all industries, demonstrating superior predictive accuracy.