Facial Feature Identification in the Deep Learning Based Apparent Personality Detection

Literature proves that the deep learning methods show significant results in apparent personality detection; however, lack of knowledge on how/why they perform well is a problem in the apparent personality detection. Plenty of explanation methods has been introduced to understand the performance of the convolutional neural network models. This work aims to describe the performance of the apparent personality detection models using Grad-CAM, Guided Backpropagation, and Guided Grad-CAM techniques. The results prove that facial features such as eyes, forehead, eyebrows, nose, and mouth are mainly involved in personality detection. However, the Guided Backpropagation method mainly highlights edges in the face area and detects background data. It was difficult for some input data with low scores to identify the facial features that impact the results. Further, the explanation methods used in this study have some limitations in describing the outputs of the apparent personality detection.