Development of a deep learning predictive model within a real-time business intelligence framework for goat milk supply sustainability

Abstract

This study aimed to develop a multidimensional data model and a prediction model for daily goat milk production within the framework of real-time business intelligence. The model was expanded to include predictions for goat milk production in Central Java, Indonesia. The Central Java daily goat milk production dataset is real data collected directly from dairy goat farms in Central Java, such as Kejar Farm in Tegal Regency, Alan’s Diary, and BK Farm in Pati Regency. The improved multidimensional data model successfully displays information hierarchically based on time, location, and farm unit dimensions through a BI dashboard. A prediction model was built by comparing the long short-term memory (LSTM) and Prophet models with several optimization scenarios. Model evaluation showed that Stacked LSTM was the best model with a 2-layer architecture. The LSTM evaluation metrics showed an RMSE value of 3.54, MAPE of 11.81%, and R 2 of 0.67, while the Prophet model was unable to capture the short-term variations and non-linear fluctuations of daily milk production data. The research results confirmed that the deep learning approach is more adaptive to the characteristics of non-stationary time series data, even in limited data. The resulting model is ready to be integrated into the RTBI Framework.