Prediksi Kepribadian Menggunakan Transfer Learning Model VGG-Face Berbasis Video

Arifin, Achmad Nurs Syururi
Nudin, Salamun Rohman

JURNAL TEKNIK INFORMATIKA DAN SISTEM INFORMASI
Vol 12 No 3 (2025): JATISI (Jurnal Teknik Informatika dan Sistem Informasi)

This study aims to develop an personality prediction system based on the Big Five Personality model using Transfer Learning with VGG-Face on video data. This research is significant as accurate personality prediction can be applied in various fields, such as behavior analysis. In this study, the pre-trained VGG-Face model, along with two LSTM layers followed by several Dense and Dropout layers, is used for facial feature extraction from video. These features are then used to predict personality across five key dimensions: openness, conscientiousness, extraversion, agreeableness, and neuroticism. The study uses secondary data from the ChaLearn Looking at People (LAP) dataset, which was utilized in the CVPR 2017 competition and includes approximately 10,000 videos. The model is evaluated using the Mean Absolute Error (MAE) metric, which is then converted into regression accuracy. The evaluation results show strong performance with accuracy: Training: 91.75%, Validation: 90.39%, and Testing: 90.28%. The results show that the model has consistency and the ability to generalize well to data it has never encountered before.