USULAN MODEL PERAMALAN PENJUALAN BERBASIS PEMBELAJARAN MESIN UNTUK PRODUK MAKE-TO-STOCK DI PT SA

• Main Author: Paramarini A P, Renata
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/83909
• Institution: Institut Teknologi Bandung

PT SA is an FMCG company facing challenges in meeting demand targets, which recently only achieved 93,4% compared to the target value of 97,5%. PT SA also deals with a high level of inventory accumulation, which reached an average of 78 days of inventory. One of the root causes of these issues is the low accuracy of PT SA's sales forecasts, which cannot serve as a reliable basis for production planning. To address this problem, this study proposes a new sales forecasting model based on a predictive machine learning model. Several alternative machine learning models are considered, including support vector machines (SVM) and variants of Recurrent Neural Network (RNN) such as Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU). These models undergo testing to identify the most suitable model for the research using k-fold cross-validation as the evaluation method. The study finds that the best-proposed model is the GRU-based model, with a total mean squared error value of 0,6966. Comparing the GRU model's forecasting results with the existing forecasting results shows an accuracy improvement in 148 out of 164 products. By categorizing the mean absolute percentage error (MAPE) values using company standards, it is found that the proportion of MAPE categorized as "Good" increases from 46% to 100%.