Prediksi Penjualan Pada Data Penjualan Perusahaan X Dengan Membandingkan Metode GRU, SVR, DAN SARIMAX

Abstract

Sales forecasting is an attempt to predict sales using several methods, such as statistical methods, machine learning, and others. Sales forecasts are considered important because poor forecasts can have an impact on the company's inventory so that it can cause storage of too much or too little goods, causing the company to lose. Therefore, we need a model that can predict sales so that companies can plan before filling stock. However, forecasts cannot be done directly, because a company's sales data is definitely influenced by various factors and sales last month are not always the same as in the future, so external data is needed in predicting future sales. Therefore, in this thesis a prediction will be made using 3 models, namely the GRU, SVR and SARIMAX models with the help of external data in the form of CPI data and inflation data. In addition, this thesis also conducted a correlation test to determine whether the sales data to be predicted has significance/relationship with external data so that it helps in predicting sales data. The results obtained from this study are that pot data is more suitable for using univariate data with the GRU model, with RMSE Train 3.22, RMSE Test 2.93. For hanger and sealware data, the best model for prediction is the SARIMAX model with univariate data type (RMSE 30.43) and multivariate data type (RMSE 8.07).