Forecasting Pet Food Item Stock using ARIMA and LSTM

Abstract

The procurement process by a company must be calculated as well as possible so that the goods or services needed by the company can be met with minimum cost. Time series forecasting is a method of predicting future events based on a set of observations in a period of time. ARIMA is a forecasting model that is used for short-term forecasting. ARIMA has some limitations with the forecast do not follow the pattern of actual series and can be applied if the data is stationary. LSTM is a modified RNN that is proposed to learn long-range dependencies across time-varying. This paper discusses forecasting one of the products that are sold on veterinary using ARIMA and LSTM so the veterinary can decide to sell the product in the future. Data that used in this forecasting were data sales of therapeutical animal food in Vet to Pet that consist of 38 data. Data splitting in this forecasting were divided into 85% of training data and 15% of testing data. Model building for forecasting was using the packages that included in Python. RMSE was used for comparing model evaluation of ARIMA and LSTM. The best method that used in forecasting therapeutical animal food was ARIMA with an RMSE value of 9.27.