Konfigurasi Hyperparameter Long Short Term Memory untuk Optimalisasi Prediksi Penjualan

Ali Khumaidi (Universitas Krisnadwipayana)
Dhistianti Mei Rahmawan Tari (Universitas Krisnadwipayana)
Nuke L. Chusna (Universitas Krisnadwipayana)

Faktor Exacta
Vol 15, No 4 (2022)

To support business development and competition, forecasting capabilities with good accuracy are required. PT. Sumber Prima Inti Motor does not want the customer's spare part needs not to be available when ordered, therefore an appropriate procurement and sales forecasting strategy is needed. Long Short Term Memory (LSTM) is a fairly good algorithm for forecasting, in this study using LSTM to predict sales of spare parts for the next 60 days. The CRISP-DM method is used and to obtain optimal model performance, hyperparameter configuration is performed. The configurations used are number of hidden layers, data partition, epoch, batch size, and dropout scenario. The best results from the LSTM model hyperparameter configuration are 3 hidden layers, 3 dropouts, epoch 150, and batch size 30. The performance of the training and testing models with RMSE is 0.0855 and 0.0846.