<![CDATA[The forecasting of highly volatile time series data remains a significant challenge due to complex, non-linear patterns. This study compared the performance of two hybrid frameworks, ARIMA-LSTM and CNN-LSTM, which were designed to integrate the statistical strengths of traditional models with the computational power of deep learning. In these architectures, the ARIMA component was utilized to extract linear trends, while the LSTM and CNN layers were employed to identify and manage non-linear dynamics within the data. Utilizing 384 monthly palm oil price data points (1993-2024) sourced from FRED, the models were evaluated using the Mean Absolute Percentage Error (MAPE) metric. The results demonstrated that the hybrid CNN-LSTM outperformed the ARIMA-LSTM and individual models, achieving a superior MAPE of 6.69%. These findings indicated that the integration of Convolutional Neural Networks (CNN) with Long Short-Term Memory (LSTM) networks was more effective in capturing the complexities of price fluctuations. Practically, the study concluded that accurate forecasting served as a critical tool for market stabilization, thereby supporting broader goals of financial certainty and ecological sustainability.]]>
Copyrights © 2026
