International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 11 Issue: 03 | Mar 2024
p-ISSN: 2395-0072
www.irjet.net
Fake Profile Detection Using Deep Learning Algorithm D. Guna Shekar, K. Siva Likith, G. Lakshmi Priya, Ms. Suvitha S, Student, Department of AI&DS, Muthayammal Engineering College, Rasipuram, Tamil Nadu, India. Student, Department of AI&DS, Muthayammal Engineering College, Rasipuram, Tamil Nadu, India. Student, Department of AI&DS, Muthayammal Engineering College, Rasipuram, Tamil Nadu, India. Assistant Professor, Department of AI&DS, Muthayammal Engineering College, Rasipuram, Tamil Nadu, India. ---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - The rise in e-scams, which can be attributed to
Understanding human interpersonal relationships can provide valuable insights into user behavior, preferences, and conversational tendencies. While these insights can enhance the quality of products and services, they can also be used against unsuspecting users. For example, online discourse can be manipulated by actors who are unknown to other participants, altering the conversation's direction.
approximately 30% of fake social media accounts, has emphasized the pressing need to identify these fraudulent profiles. Due to the limitations of the current model in handling multi-model networks, efforts have been made to address real-time issues. This research introduces an advanced deep-transfer learning model that enhances the detection of fake profiles by conducting a comprehensive analysis of diverse social media data samples. Our model collects a wide range of data from various social media platforms, including posts, likes, comments, multimedia content, user activity, login behaviors, and more. Each type of data is individually processed to identify suspicious patterns that are indicative of fake accounts. For example, discrepancies like male profiles predominantly posting about or using images of females. Similarly, audio signals undergo transformations such as 1D Fourier, Cosine, Convolutional, Gabor, and Wavelet Transforms. On the other hand, image and video data are processed using their 2D counterparts. Text data is transformed using Word2Vec, which assists our binary Convolutional Neural Network (bCNN) in distinguishing between genuine and fake profiles. Feature optimization is carried out using the Grey Wolf Optimizer (GWO) for 2D data and the Elephant Herding Optimizer (EHO) for 1D data, ensuring minimal redundancy in features. Subsequently, separate 1D CNN classifiers are employed to classify the refined features and identify fake profiles. The results from these classifiers are combined through a boosting mechanism. Our findings demonstrate an increase in accuracy by 8.3%, precision by 5.9%, and recall by 6.5% compared to traditional methods.
The anonymity provided by SMPs is their main attraction, but it can also be their downfall. Users can create fake profiles, causing distress to unsuspecting victims. Activities such as rumor-mongering, cyberbullying, and the dissemination of false information are examples of the negative use of these platforms. The competitive nature of SMPs, as demonstrated by metrics like "likes" or "followings," exacerbates the issue, prompting users to use both overt and covert tactics to gain an advantage.
1.1 Motivation and Contributions The digital revolution has brought about a new era of global connectivity and information sharing. However, alongside these positive developments, there is a darker side that cannot be ignored. Recent estimates indicate that nearly one-third of all social media accounts are fake, highlighting the growing threat of e-scams and of news, communication, and business interactions for many individuals. The presence of fraudulent entities on social media undermines trust and sabotages genuine interactions, creating an environment that is ripe for cybercrimes. Unfortunately, the existing models designed to address this issue often fall short. They are either ineffective, unable to handle multimodal data, or too complex to be applied in real-time situations. As social media continues to play an increasingly significant role in personal, professional, and societal spheres, there is an urgent need for an advanced solution that prioritizes user safety, trust, and platform integrity.
Key Words: Social Media, Fake, Profile GWO, EHO, CNN, Multimodal, Cold Start, Issues.
1.INTRODUCTION The rise of big data platforms, particularly social media, has brought about a new set of challenges, including identity theft, due to their widespread popularity. Unfortunately, spammers and con artists have taken advantage of these platforms, leading to an increase in cybercrime activities such as spamming. The impact of these threats on Social Media Platforms (SMPs) varies, but their existence cannot be denied.
© 2024, IRJET
|
Impact Factor value: 8.226
2. METHODOLOGY The model in this study utilized various techniques such as XG Boost, a random forest method, and a profile-focused multi-layered neural network. These techniques were employed to extract observable features from the data, which were then saved in a CSV file for easy reading by the
|
ISO 9001:2008 Certified Journal
|
Page 437