PERANCANGAN MODEL MACHINE LEARNING UNTUK DETEKSI ANOMALI DAN PREDIKSI UNTUK PARAMETER PENJADWALAN PRODUKSI DI PT X

• Main Author: Fijar Pradana, Raditya
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/62094
• Institution: Institut Teknologi Bandung

PT X, one of the largest automotive company in Indonesia, has a market share of 31.3% in Indonesia in the period from January 2021 to July 2021. In addition, PT X's contribution to the export market is quite large. To maintain and increase this market share, PT X needs to maintain customer satisfaction by providing good after sales service, especially the procurement of service parts. The types of service parts that need to be provided by PT X are very diverse with very varied characteristics, such as the fluctuating number of orders and lead times. To carry out the production process of the service part, it is necessary to have good production planning and control. Therefore, anomaly detection models and prediction or forecasting models were built. The anomaly model becomes a trigger for the prediction model to run. The prediction model will be used as input in the revision of forecasting results, verification of production capacity, and rescheduling. Anomaly detection and prediction models will be built using data mining and machine learning techniques, using the Cross Industry Standard Process for Data Mining (CRISP-DM) methodology. Anomaly detection model is built using the LSTM autoencoder algorithm and control chart. The prediction model is built using the LSTM autoencoder algorithm and simple moving average. In detecting anomalies and prediction process, the two models used have quite different characteristics. In general, the LSTM autoencoder is a more robust model. About the anomalies, of the eleven service parts analyzed, only eight service parts have anomalies. Furthermore, the detected anomaly is followed up by the prediction model and recorded in resource isolation.