머신러닝 기법을 활용한 서비스 산업의 일단위 수요예측 연구

초록

Purpose – As the volume of data has increased due to advancements in supply chain management and IT technology, there is growing interest in demand forecasting using various types of time series data. This paper studies daily demand forecasting by comparing traditional time series prediction methods with machine learning-based prediction techniques to improve forecasting accuracy. Design/Methodology/Approach – Using daily sales data from Bakery A’s hierarchy ERP system, a Hidden Markov Model (HMM) algorithm was applied to estimate the number of hidden state and perform daily demand forecasting. The accuracy of the HMM-based forecasts was then compared and analyzed against existing time series prediction methods. Findings – The application of the Hidden Markov Model (HMM) at the end-item level demonstrated higher forecasting accuracy when compared to traditional time series methods across all forecast horizons. HMM proved to be highly effective in maintaining both efficiency and accuracy in daily demand forecasting, making it a suitable method for companies that require daily operational planning. Research Implications – This study reviews existing research on time series prediction methods and introduces a new machine learning-based approach. By proposing an enhanced Hidden Markov Model, this research offers a method that can predict future daily demand by inferring state changes within data. Further research is needed to explore daily demand forecasting models in similar industries or service sectors, with a focus on comparing and evaluating applicability.

키워드