loading page

Deepfake Detection with Choquet Fuzzy Integral
  • İsmail İLHAN,
  • Mehmet Karaköse,
  • Serhat ATAŞ
İsmail İLHAN
Adiyaman University

Corresponding Author:[email protected]

Author Profile
Mehmet Karaköse
Firat University
Author Profile
Serhat ATAŞ
Firat University
Author Profile


Deep forgery has been spreading quite quickly in recent years and continues to develop. The development of deep forgery has been used in films. This development and spread have begun to concern people about security, as well as being a threat to most companies and statesmen. Although deepfake videos are used for humor, they have also been used for malicious purposes. In this field, businessmen have been blackmailed that their speeches have been made in different ways by imitating the images of statesmen. Deep fraud detection procedures are carried out to prevent this threat. Deep forgery has outpaced deepfake detection processes. That’s why most platforms and companies have supported developers to help struggle deepfakes and rewarded them. This study is conducted to contribute this struggle, 3 different deepfake algorithms using Mesonet, Resnet and EfficientNet methods are used. Moreover, a new deepfake detection method is presented by combining them with Choquet Fuzzy Integral. The method we have proposed has taken 3 different algorithms that are good in their fields and collected the accuracy values that each algorithm can work alone, the fuzzy membership values under a single roof using Choquet Fuzzy Integral and thus has significantly increased the accuracy rate of deepfake detection by signing a study that has not been done before in the deepfake field. One of the contributions of the method we have proposed is to combine the algorithms that are trained in different data sets and detect in different ways and to use the areas where these algorithms are good in a single method. Experimental results using FaceForensics++, DFDC, Celeb-DF-v2 and DeepFake-TIMIT-HQ dataset show that the proposed approach based on Choquet fuzzy integral technique for deepfake classification outperforms single classifiers and achieves the highest accuracy of 97%. In this method, a more effective result can be achieved by using other effective models. More algorithms can be used in the method or can be replaced with new proposed method. We believe that the proposed method will inspire researchers and be further improved.