Hybrid of Neural Prophet and Attention-Based LSTM for Container Freight Rate Forecasting

Time series forecasting has been widely employed in diverse industries, including finance, retail, healthcare, and maritime forecasting. Achieving accurate forecasts in these applications, particularly in the maritime industry, is crucial but challenging. Due to its superiority over a single model, hybridization between many models has begun to develop to enhance model performance. This study introduces a new hybrid forecasting approach that combines the recently developed Neural Prophet (NP) model and the Attention-Based Long-Short-Term-Memory (LSTM) model. The proposed method is designed to predict container freight rates. This proposed model will be compared, discussed, and evaluated with different performance metrics later. The result found that the proposed hybrid model shows the good performance of least performance metrics of MAE and MSE values of 0.0261 and 0.00087 for CCFI, 0.165 and 0.037 for SCFI compared to a single model of Neural Prophet. For MAPE, only the Drewry dataset in the proposed method has the least value of 0.028 compared to a single model of Neural Prophet.