Classification of Personality Traits by Using Pretrained Deep Learning Models

Abstract

Nowadays, personality traits analysis has become one of the important things since international companies need to hire employees and be used in education and forensic verification. In this paper, three pre-trained models of deep learning were evaluated to classify an individual's personality traits from his signature after analyzing, processing, and labeling the data and dividing it into five categories according to the Big Five factor. The analysis is based on 6600 images divided into three groups (training, testing, and validation). Data Augmentation was used to overcome the lack of data and its imbalance. Also, transfer learning was used that represented by the three models (VGG16, Inception, and ResNet50), which work on the principle of freezing the first layers and updating the last layers to take advantage of the pre-trained weights and obtain the lowest error rate. Results showed that the ResNet-50 achieved the best classification accuracy with up to 99% and the lowest error rate with 0.0304. While the InceptionV3 model outperformed VGG16 in the training phase of 99%, but in the validation phase, the VGG16 provided the Highest accuracy of 98% and the least error of 0.1090.