Development of a hybrid ARIMA and K-Means approach for traditional medicine production forecasting

Vollständiger Abstract

Worum geht es in dieser Arbeit?

The object of research is traditional medicine as important component of Indonesia’s cultural heritage and national identity. Traditional medicine known as jamu is one of Indonesia's cultural heritages that has become a national identity. Each region in Indonesia has its own distinctive characteristics, such as Madurese jamu, which marketing has expanded internationally. The widespread marketing requires the small and medium enterprises of traditional medicine to maintain its production. One way to ensure that traditional herbal medicine production is maintained is by predicting future production. In predicting traditional medicine, several components are needed that can be used over a certain period of time, such as inventory and previous sales. One prediction method that can be used is Autoregressive Integrated Moving Average (ARIMA), with the stages of parameter identification, parameter estimation, model verification, and prediction. However, there are several problems with the dataset, namely irrelevant features and outlier data. Therefore, statistical significance is used in feature selection using Pearson correlation and Variance Inflation Factor (VIF), as well as K-Means Clustering for outlier detection. From the several scenarios conducted, the results show that the ARIMA method with statistical significance and K-Means Clustering has the lowest MAPE of 10.252%. Meanwhile, ARIMA with statistical significance has a MAPE of 21.264%, ARIMA with K-Means Clustering has a MAPE of 31.263%, and ARIMA alone has a MAPE of 38.558%. From the research shows that the ARIMA combination with statistical significance and K-Means Clustering is able to provide better performance in predicting traditional medicine production.