Optimalisasi Algoritma Random Forest Menggunakan SMOTE untuk Prediksi Pembatalan Tamu Hotel

Abstract

The role of technology has simplified user access to various services, including hotel reservations. Technology has also made it easier for service providers to offer their products by showcasing hotel photos and specifications. However, the ease of making reservations aligns with the ease for users to cancel their hotel room bookings, which undoubtedly has a negative impact on hotel owners. To minimize losses, cancellation penalties and predictions regarding the likelihood of reservation cancellations can be implemented. Therefore, predicting hotel reservation cancellations has become crucial for hotel managers to reduce occupancy declines. One of the tools for prediction can be done through machine learning technology. The researcher utilized a hotel reservation dataset consisting of 18 attributes, which underwent data processing using the Random Forest algorithm, where the label attribute was the decision to cancel or not. In this study, the Random Forest algorithm was used without optimization, yielding an accuracy of 87.85%. Subsequently, the dataset, before being processed using Random Forest, underwent handling for imbalanced data using SMOTE. Through this method, the accuracy was enhanced by 2.46%, resulting in a final accuracy of 90.31%. Since the classification was only divided into 2 groups, SMOTE was performed only once to obtain a balanced data sample. This research contributes significantly to understanding the use of machine learning algorithms, particularly Random Forest, in addressing the challenges of imbalanced data in the case of predicting hotel reservation cancellations. The research highlights the potential of SMOTE as an effective tool in handling data imbalance, which can assist hotel managers and hospitality service providers in optimizing reservation management and improving room occupancy rates.