Objective: A wide range of social consequences is attributed to face preferences, from mate choices and decisions about social relationships to hiring decisions, including fashion modeling and marketing issues. This paper aims to design and develop an expert system based on the fuzzy inference system (FIS) for ranking facial attractiveness. Methods: Firstly, we reviewed the research on facial attractiveness and found that most papers do not use an expert system to rate the facial attractiveness of people. Therefore, this study used a deep learning method using a convolutional neural network (CNN) to recognize attractive faces. Then, with the help of the rules of the fuzzy inference system, an expert model was designed for ranking facial attractiveness. Results: We demonstrated that the combination of FIS and CNN is highly effective and excellent at ranking facial attractiveness. Our method performs better than other methods we have investigated in a small amount of data. The mean and standard deviation values of sensitivity, specificity, precision, and accuracy of the proposed model for detecting attractive and unattractive faces were 99.46±0.03%, 99.35±0.01%, 98.99±0.02%, and 99.7±0.01%, respectively. In addition, we have obtained a mean prediction accuracy for ranking facial attractiveness as 99.53+0.06%. Conclusion: We designed an expert ranking system for facial attractiveness to rank different faces according to facial structure features. As a result of our approach, we could also provide a deeper understanding of how FIS and CNN can recognize ranking facial attractiveness.

Objective: A wide range of social consequences is attributed to face preferences, from mate choices and decisions about social relationships to hiring decisions, including fashion modeling and marketing issues. This paper aims to design and develop an expert system based on the fuzzy inference system (FIS) for ranking facial attractiveness. Methods: Firstly, we reviewed the research on facial attractiveness and found that most papers do not use an expert system to rate the facial attractiveness of people. Therefore, this study used a deep learning method using a convolutional neural network (CNN) to recognize attractive faces. Then, with the help of the rules of the fuzzy inference system, an expert model was designed for ranking facial attractiveness. Results: We demonstrated that the combination of FIS and CNN is highly effective and excellent at ranking facial attractiveness. Our method performs better than other methods we have investigated in a small amount of data. The mean and standard deviation values of sensitivity, specificity, precision, and accuracy of the proposed model for detecting attractive and unattractive faces were 99.46±0.03%, 99.35±0.01%, 98.99±0.02%, and 99.7±0.01%, respectively. In addition, we have obtained a mean prediction accuracy for ranking facial attractiveness as 99.53+0.06%. Conclusion: We designed an expert ranking system for facial attractiveness to rank different faces according to facial structure features. As a result of our approach, we could also provide a deeper understanding of how FIS and CNN can recognize ranking facial attractiveness.

Objective: A wide range of social consequences is attributed to face preferences, from mate choices and decisions about social relationships to hiring decisions, including fashion modeling and marketing issues. This paper aims to design and develop an expert system based on the fuzzy inference system (FIS) for ranking facial attractiveness. Methods: Firstly, we reviewed the research on facial attractiveness and found that most papers do not use an expert system to rate the facial attractiveness of people. Therefore, this study used a deep learning method using a convolutional neural network (CNN) to recognize attractive faces. Then, with the help of the rules of the fuzzy inference system, an expert model was designed for ranking facial attractiveness. Results: We demonstrated that the combination of FIS and CNN is highly effective and excellent at ranking facial attractiveness. Our method performs better than other methods we have investigated in a small amount of data. The mean and standard deviation values of sensitivity, specificity, precision, and accuracy of the proposed model for detecting attractive and unattractive faces were 99.46±0.03%, 99.35±0.01%, 98.99±0.02%, and 99.7±0.01%, respectively. In addition, we have obtained a mean prediction accuracy for ranking facial attractiveness as 99.53+0.06%. Conclusion: We designed an expert ranking system for facial attractiveness to rank different faces according to facial structure features. As a result of our approach, we could also provide a deeper understanding of how FIS and CNN can recognize ranking facial attractiveness.