Forecasting Patchouli (Pogostemon Cabin Benth) Production in North Kolaka Using ARIMA, LSTM and Hybrid ARIMA-LSTM

Bakhtiar, Sri Muslihah (Actuarial Science Study Program, Universitas Muhammadiyah Kolaka Utara)
Fajar, Nurhikmah (Computer Engineering Study Program, Universitas Muhammadiyah Kolaka Utara)

Zero : Jurnal Sains, Matematika, dan Terapan
Vol 9, No 2 (2025): Zero: Jurnal Sains Matematika dan Terapan

Patchouli (Pogostemon cablin Benth.) is one of Indonesia’s most economically valuable essential oil commodities. This study forecasts patchouli production using ARIMA, LSTM, and a hybrid ARIMA–LSTM model, emphasizing the novelty of applying machine learning techniques to essential oil production forecasting, an understudied area. Weekly production data from ARS Atsiri North Kolaka (January 2022–July 2025, 187 records) were analyzed. ARIMA was applied to capture linear patterns, LSTM to model nonlinear dynamics, and the hybrid model to combine both characteristics. Model performance was evaluated using MSE, RMSE, and MAPE. The ARIMA (2,1,1) model performed best among the linear approaches, while LSTM with normalization, windowing, and 50 hidden units achieved the highest overall accuracy (MSE = 251.22, RMSE = 321.67, MAPE = 0.238%). The hybrid model did not outperform LSTM, likely due to the limited dataset and the dominance of nonlinear patterns, thus confirming LSTM as the most effective model.