Sistem Deteksi Kepuasan Pelanggan dengan Teknik Pengelolaan Citra Menggunakan Convolutional Neural Networks

Abstract

Advancements in computer vision and facial expression recognition provide a new, objective, and non-intrusive method for measuring customer satisfaction in real time. This study develops a customer satisfaction detection system at Rumah Diskusi ALCO Café using Convolutional Neural Networks (CNN) with a mixed-methods approach, combining quantitative and qualitative analysis. The RAF-DB dataset containing 15,339 labeled images (12,271 for training and 3,068 for testing) across seven emotion classes was processed through image acquisition, preprocessing, and ResNet50 fine-tuning. The resulting model achieved an accuracy of 80.34%, with a Precision of 83.55%, Recall of 81.78%, and F1-Score of 82.32% on the test data. Field implementation over four weeks successfully recorded and analyzed thousands of customer facial expressions in key areas such as the cashier and main seating area in real time. Results showed a customer satisfaction distribution of approximately 72% “Satisfied,” 16% “Quite Satisfied,” and 12% “Not Satisfied,” with a declining trend during peak hours in the afternoon. Cross-validation with customer surveys demonstrated a strong correlation between the system’s predictions and reported satisfaction, proving the effectiveness of this method as a real-time monitoring tool. The study contributes a practical technical and methodological framework that can be replicated in other service industries for objective and real-time customer satisfaction monitoring.