HYPERPARAMETER TUNING PADA MODEL PREDIKSI WAKTU TUNGGU ANTRIAN BERBASIS REGRESI RANDOM FOREST UNTUK MENINGKATKAN AKURASI MODEL STUDI KASUS: PUSAT LAYANAN PELANGGAN DI JAKARTA

• Main Author: Halim, Stevani
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/48352
• Institution: Institut Teknologi Bandung

Queue is an event caused by the needs for service that exceed the service capacity and is commonly found in daily life at the locations of service provider, one of which is the customer service center. The queues at the customer service center in Jakarta which is the location of the author's case study are now well managed with the use of a queue management system. This queue management system is equipped with the ability to display queuing status information to customers, so customers know the estimated waiting time for each of their queue number. Even so, the prediction accuracy of the queue waiting time displayed to customers is still fairly low because it is only calculated based on the average queue waiting time of the last 3-5 queue numbers which were called. To solve the problem above, the author optimizes the queue waiting time prediction model based on random forest regression. This random forest regression algorithm was chosen because it has the best performance among other regression-based machine learning algorithms, for example linear regression. Optimization is done using one of the hyperparameter tuning methods, which is grid search. The methodology used to optimize this prediction model is a CRISP-DM methodology with a slight modification. The stages of this methodology begin with business understanding, data acquisition and understanding, data preparation, baseline definition, model optimization, and end with evaluation. The result obtained from this final project is a queue waiting time prediction model based on random forest regression that has been optimized using hyperparameter tuning. The best combination of hyperparameters obtained is n_estimators = 100 and random_state = 25. The accuracy of this model is 93.77%, a significant increase of 2.47% after hyperparameter tuning.