InstaGuard: An AI‑Powered Framework for Detecting Fraudulent

Activities on Instagram Using Machine Learning and Deep Learning

Techniques

Palani Murugan S.

, Gobinath R.

, Rahapriya K.

, Fahumitha Afrose B.

Fathima Fazlina M.

and Durga Devi S.

Department of AI&DS, E.G.S.Pillay Engineering College (Autonomous), Nagapattinam, Tamil Nadu, India

E.G.S. Pillay Engineering College (Autonomous), Nagapattinam, Tamil Nadu, India

Keywords: Fraud Detection, Social Media Security, Instagram Fraud, Machine Learning, Deep Learning, Natural

Language Processing (NLP), Fake Account Detection, Phishing Prevention, Misinformation Detection,

Cybersecurity, Real‑Time detection.

Abstract: The rapid growth of Instagram has raised major security concerns, including fake profiles, deceptive

promotions, phishing scams, impersonation, and misinformation. These threats mislead users, exploit

audiences, and increase cyber risks. To address these challenges, we propose InstaGuard, an AI-powered

fraud detection system for Instagram. InstaGuard employs advanced AI techniques to analyze user behavior

and content. Random Forest and XGBoost detect fake accounts based on profile attributes and engagement

trends. BERT-based NLP models classify misleading promotions and scams, while LSTM networks and URL

reputation verification identify phishing attempts. CNNs and Siamese Networks handle impersonation

detection, while RoBERTa-based transformers and graph-based content analysis mitigate misinformation. By

integrating Instagram’s Graph API for automated data collection, InstaGuard enables real-time fraud detection

and response. Simulations indicate up to 95% accuracy, reducing false positives by 30% compared to existing

methods. Its real-time capabilities ensure swift action against fraudulent activities, enhancing platform

security.This research contributes to AI-driven cybersecurity by introducing a scalable, adaptive fraud

detection framework tailored for Instagram. InstaGuard strengthens user safety by mitigating fraud and

misinformation, reinforcing platform integrity in an increasingly complex digital landscape.

1 INTRODUCTION

Current approaches to identifying fraudulent activity

face significant limitations. Conventional methods

such as manual content review, blocklist

implementations, and keyword filtering systems

demonstrate inadequate responsiveness to constantly

evolving deceptive practices. The inherent latency in

human reporting mechanisms and the inability of

blocklist solutions to recognize novel fraudulent

patterns create substantial detection gaps.

Furthermore, content moderation teams encounter

operational challenges when processing the

exponentially growing volume of suspicious material.

To overcome these limitations, we developed

InstaGuard - an artificial intelligence-powered

detection framework specifically designed for

Instagram. The system integrates multiple advanced

machine learning components including Random

Forest and XGBoost classifiers for anomalous

activity identification, BERT-based natural language

processing architectures for deceptive promotional

content recognition, and LSTM neural networks

combined with URL heuristic analysis for phishing

attempt detection. For profile impersonation cases,

the platform utilizes convolutional neural networks

and Siamese network architectures to perform visual

similarity assessments. The system additionally

incorporates RoBERTa transformer models and

graph-based analytics for misinformation

identification. Through seamless integration with

Instagram's Graph API, the solution provides real-

time monitoring capabilities while maintaining strict

compliance with platform regulations. Experimental

results indicate the system achieves 95% detection

840

S., P. M., R., G., K., R., B., F. A., M., F. F. and S., D. D.

InstaGuard: An AI-Powered Framework for Detecting Fraudulent Activities on Instagram Using Machine Learning and Deep Learning Techniques.

DOI: 10.5220/0013944600004919

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 5, pages

840-847

ISBN: 978-989-758-777-1

accuracy with a 30% reduction in false positive

incidents compared to existing solutions. The

platform's dynamic risk scoring mechanism enables

efficient prioritization of suspicious cases,

representing a scalable approach to fraudulent

activity prevention. The architecture's adaptive

learning capabilities ensure continued effectiveness

against emerging threat vectors.

2 RELATED WORK

2.1 Profile-Based Detection

The profile analysis subsystem examines multiple

account characteristics including temporal metrics

(account age, activity patterns), social graph

properties, content posting regularity, and

engagement quality indicators. This approach

surpasses traditional detection methodologies

employing support vector machines and logistic

regression through its hybrid architecture combining

ensemble methods with manual verification

protocols.

2.2 Content-Based Detection

The content analysis module processes textual,

visual, and video data through transformer-based

natural language processing pipelines for linguistic

deception detection, computer vision architectures for

manipulated media identification, and adversarial

example recognition systems for robust classification.

This hierarchical analysis structure enables

identification of sophisticated fraudulent content that

evades conventional pattern matching techniques.

InstaGuard overcomes these limitations by utilizing

BERT-based NLP models to analyze deceptive text

patterns and detect scam-related language.

Additionally, deepfake detection algorithms,

models such as Convolutional Neural Networks

(CNNs) and Generative Adversarial Networks

(GANs), InstaGuard authenticates images effectively.

The integration of synthetic media recognition and

adversarial training enhances its ability to detect

content-based fraud with greater accuracy.

2.3 Network-Based Detection

Network-based detection focuses on analyzing user

interactions—such as follower networks, comment

behaviors, and shared links to identify coordinated

bot networks and misinformation campaigns. Graph-

based machine learning techniques, including Graph

Neural Networks (GNNs), have been employed to

detect fraudulent interactions in large-scale social

networks (Jiang et al., 2022). However, scalability

remains a major challenge, as analyzing massive

social graphs demands high computational resources

and often leads to delays in real-time detection.

InstaGuard addresses this limitation by

implementing optimized GNN models, which

efficiently analyze large-scale social graphs and

detect coordinated fake engagement patterns with

reduced computational overhead. Moreover,

InstaGuard leverages Instagram’s Graph API for real-

time fraud monitoring, allowing instant intervention

against suspicious activities.

2.4 Comparison & InstaGuard’s

Contributions

Existing fraud detection techniques struggle with

three primary challenges. First, evasion tactics, where

bots and scammers increasingly mimic human

behavior, making profile-based detection less

reliable. Second, contextual limitations, as content-

based models lack deep contextual understanding,

making them vulnerable to adversarial modifications.

Finally, scalability issues, since network-based

approaches require high computational resources,

limiting their real-time effectiveness.

InstaGuard introduces an integrated AI-driven

approach that combines profile analysis, content

verification, fake link detection, and network analysis

to overcome these limitations. Preliminary

experiments on a diverse Instagram dataset

demonstrate that InstaGuard achieves a 15% higher

accuracy in detecting fake accounts compared to

traditional profile-based methods. Moreover, its real-

time capabilities and fraud risk scoring system enable

proactive intervention, significantly reducing

fraudulent activity on the platform.

3 PROPOSED METHODOLOGY

Detecting fraudulent activities on social media

platforms has become increasingly complex.

Traditional methods such as manual moderation, rule-

based filtering, and user reports are often insufficient

in combating evolving fraud tactics. To address this,

InstaGuard is proposed as an AI-powered fraud

detection system designed specifically for Instagram.

The system processes Instagram profile data, post

content, and external links to detect fraudulent

behaviors. The dataset used for training and

InstaGuard: An AI-Powered Framework for Detecting Fraudulent Activities on Instagram Using Machine Learning and Deep Learning

Techniques

841

evaluation includes real-world Instagram data,

ensuring adaptability and high accuracy. The fraud

detection process follows a structured approach, as

described below.

3.1 Data Collection Using Instagram

Graph API

The Instagram Graph API is used to retrieve real-time

data from Instagram. The system collects profile data,

including username, account type, follower/following

count, and account creation date. Additionally, post

data such as captions, hashtags, image and video

URLs, and timestamps are extracted. Engagement

metrics, including like count, comment count, and

follower growth trends, are also gathered. External

links embedded in bios and posts, including shortened

URLs, are collected for further analysis. The API

ensures ethical and efficient data collection while

complying with Instagram’s data access policies.

To ensure dataset diversity and

representativeness, data is collected from a wide

range of accounts, including personal, business, and

influencer profiles. Fraudulent accounts are labeled

using a combination of manual annotation, automated

labeling tools, and crowdsourcing approaches.

Manual annotation involves experts analyzing user

behavior and engagement patterns, while automated

AI-based fraud detection tools help validate

suspicious accounts. Additionally, human annotators

are engaged in the labeling process to improve

accuracy and reduce bias.

3.2 Feature Engineering

To enhance model performance, key fraud indicators

are extracted. Engagement-based features include

follower-to-following ratio, comment sentiment

analysis, and detection of sudden engagement spikes.

Temporal features such as account age, posting

frequency, and behavioral patterns help distinguish

fraudulent activities from legitimate ones. Content-

based features focus on NLP-based text analysis to

identify scam-related language and computer vision

techniques to detect anomalies in images.

Handling data imbalance is a crucial step in fraud

detection. To address this issue, the dataset is

balanced using the Synthetic Minority Oversampling

Technique (SMOTE). Additionally, weighted loss

functions are incorporated to ensure the model does

not favor legitimate accounts over fraudulent ones.

3.3 Model Selection and Training

InstaGuard employs supervised Various machine

learning algorithms, including Decision Trees,

Random Forest, and Deep Neural Networks (DNNs),

are utilized for fraud classification. Decision Trees

and Random Forest models efficiently handle

structured data, whereas Deep Neural Networks are

adept at identifying intricate fraud patterns in large

datasets. XGBoost is included for its high accuracy

and ability to handle imbalanced data efficiently.

To improve transparency and interpretability,

explainability techniques such as to enhance model

interpretability, SHAP (SHapley Additive

exPlanations) and LIME (Local Interpretable Model-

agnostic Explanations) are incorporated. SHAP

provides insights into feature importance, helping to

understand which factors contribute most to fraud

detection, while LIME explains individual fraud

predictions to ensure the model’s reliability.

3.4 Phishing Link Detection

Detecting phishing links is a crucial component of

fraud prevention. InstaGuard implements URL

analysis using pre-trained models and external threat

intelligence databases. A real-time URL scanning

mechanism is integrated with databases such as

VirusTotal to identify and flag malicious links.

Additionally, shortened URLs are expanded to their

original form to ensure accurate analysis.

3.5 Evaluation and Optimization

The model is fine-tuned leveraging essential

performance indicators such as precision, recall, and

F1-score to ensure high fraud detection accuracy.

Cross-validation techniques, such as k-fold

validation, are used to assess the model's robustness

across different data subsets. Furthermore, an error

analysis process is conducted to analyze instances of

false positives and false negatives, allowing for

continuous improvements in model performance.

3.6 Deployment and User Interface

InstaGuard is deployed as a web-based tool or API to

detect fraud in real-time. The system is designed for

scalability, utilizing cloud-based platforms such as

AWS and Google Cloud to efficiently process large

volumes of data with minimal latency. Security

measures, including data anonymization and

encryption, are implemented to ensure user privacy

during data processing.

To enhance usability, an interactive dashboard is

developed. This dashboard provides fraud risk scores,

assigning a risk level to each user or post based on

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

842

fraudulent activity likelihood. Real-time alerts notify

users about potentially fraudulent activities, enabling

timely intervention. Additionally, detailed reports

offer insights into flagged accounts, suspicious

behavior patterns, and overall fraud trends.

Figure 1: Flowchart of proposed solution.

4 IMPLEMENTATION &

DATASETS

4.1

Implementation

of InstaGuard

InstaGuard is an AI-powered framework designed to

detect fraudulent activities on Instagram using real-

time data collected via

Instagram’s Graph API. The system utilizes

machine learning (ML) and deep learning (DL)

techniques to examine fraud patterns, focusing on

fake accounts, scams, phishing, impersonation, and

misinformation. With a targeted overall accuracy of

95%, InstaGuard continuously improves its models

by optimizing feature extraction, refining training

techniques, and integrating real-world data updates.

4.1.1 Real-Time Data Collection

To ensure highly accurate fraud detection, InstaGuard

exclusively collects real-time data from Instagram’s

Graph API. This API enables the system to retrieve

user profile details such as username, bio, profile

picture, and verification status. Engagement metrics,

including followers, following count, likes,

comments, and story views, are also gathered.

Additionally, post attributes like captions, hashtags,

URLs, and media types, along with interaction

behavior such as message patterns and reply rates, are

extracted. All collected data is stored in a structured

format, ensuring efficient retrieval and processing.

Unlike static datasets, this real-time approach allows

InstaGuard to dynamically adapt to emerging fraud

patterns, reducing outdated model biases and

improving overall detection accuracy.

4.1.2 Data Preprocessing & Feature

Engineering

Since real-time data can contain noise and

inconsistencies, InstaGuard employs a robust

preprocessing pipeline. Data cleaning removes

missing values, duplicate records, and irrelevant data,

ensuring high-quality input for the models. Feature

engineering extracts advanced fraud-related patterns,

such as engagement consistency and suspicious URL

behavior. Text data undergoes tokenization, stopword

removal, and transformation using TF-IDF and BERT

embeddings to enhance NLP-based fraud detection.

Image preprocessing involves resizing, grayscale

conversion, and CNN-based feature extraction for

detecting impersonation and visual fraud patterns.

Numerical data, such as follower count and

engagement ratios, is normalized to maintain fair

comparisons across accounts. Continuous feature

refinement is performed based on real-time fraud

trends detected through Graph API data streams,

ensuring the models remain adaptive and effective.

4.1.3 Machine Learning & Deep Learning

Models

InstaGuard integrates specialized ML and DL models

for each fraud category, continuously refining them

to achieve 95% accuracy. Fake account detection is

performed using XGBoost and Random Forest,

analyzing engagement anomalies and account

behaviors. With optimized feature selection, the

accuracy has been improved from 70% to 92%. Scam

detection is implemented using BERT and RoBERTa,

classifying deceptive promotional content with an

accuracy of 93%. Phishing link detection is achieved

using LSTM networks and URL reputation models,

reaching 95% accuracy by integrating real-time

domain validation and SSL verification.

Impersonation detection employs CNN and Siamese

Networks, refining facial similarity analysis and

improving detection accuracy from 72% to 94%.

Misinformation detection is enhanced through graph-

based NLP models with RoBERTa, flagging

InstaGuard: An AI-Powered Framework for Detecting Fraudulent Activities on Instagram Using Machine Learning and Deep Learning

Techniques

843

misleading content with 95% accuracy by

incorporating fact-checking APIs. Each model

undergoes continuous retraining using real-time data

to ensure adaptability against evolving fraud tactics.

4.1.4 Fraud Score Calculation & Real-Time

Alerts

To provide precise fraud detection, InstaGuard

assigns a Fraud Probability Score (FPS) to every

analyzed profile, post, or link. This score is computed

based on three key metrics: Profile Risk Score (PRS),

which measures suspicious user behavior; Content

Trust Score (CTS), which evaluates post credibility

using NLP techniques; and Image Similarity Index

(ISI), which detects impersonation using deep

learning models. When the fraud score exceeds 75%,

real-time alerts are triggered, notifying Instagram

moderators for verification. To enhance response

efficiency, InstaGuard supports automated fraud

intervention mechanisms, warning users about

potential scams before interaction.

4.1.5 Ethical Considerations & Scalability

Since InstaGuard collects real-time data, privacy

compliance is strictly maintained by processing only

publicly available data. To enhance scalability,

models are optimized using distributed cloud

computing, ensuring minimal latency during fraud

detection. Furthermore, Techniques from explainable

AI (XAI), including SHAP and LIME, enhance

transparency in fraud classification, reducing false

positives and ensuring trust in the system’s decisions.

4.1 Dataset Description

Unlike traditional fraud detection models that rely on

static datasets, InstaGuard is trained exclusively on

real-time data from Instagram’s Graph API. This

ensures the system remains adaptive to new fraud

patterns and eliminates dataset obsolescence. The

dataset includes real-time fake and legitimate

accounts extracted directly from Instagram using

profile behavior analysis. Scam advertisement data is

continuously updated with newly detected fraudulent

promotions. Phishing URLs are sourced from active

Instagram scam reports and cybersecurity databases.

Impersonation data is derived from real-time profile

comparisons with verified accounts, and

misinformation data is validated through integrations

with fact-checking sources. Since InstaGuard

dynamically collects new fraud cases, model

retraining is performed continuously to improve

detection accuracy.

Data Preprocessing Enhancements for Higher

Accuracy To achieve 95% fraud detection accuracy,

InstaGuard incorporates advanced preprocessing

techniques. Text data processing is enhanced using

BERT embeddings and attention mechanisms for

improved NLP classification. Image data is refined

with CNN training and adversarial augmentation,

boosting impersonation detection. URL data

undergoes real-time phishing link classification by

tracking domain age and SSL verification.

Additionally, active learning techniques are

implemented, allowing the models to refine

predictions based on real-world user reports and

moderator feedback.

5 RESULT AND DISCUSSION

The evaluation of InstaGuard’s fraud detection

system was conducted through extensive

experimentation, performance benchmarking, and

comparative analysis with existing fraud detection

techniques. The results indicate significant

enhancements in accuracy, precision, recall, and F1-

score across different fraud categories. This section

provides an in-depth discussion of the effectiveness

of the proposed methodology, the impact of real-time

data collection, and the implications for enhancing

fraud detection on social media platforms.

Figure 2 shows the Performance Comparison of

Machine Learning Models for Fraud Detection.

Figure 2: Performance comparison of machine learning

models for fraud detection.

5.1 Performance Evaluation

To assess the reliability of InstaGuard, multiple

machine learning and deep learning models were

trained, tested, and validated using realtime

Instagram data. The dataset used for evaluation

contained over 500,000 Instagram profiles and 10

million posts Fraud detection effectiveness was

assessed using key metrics such as accuracy,

precision, recall, and F1-score. These metrics

provided a comprehensive evaluation of system

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

844

performance, ensuring reliable identification of

fraudulent activities across various scenarios and

datasets, while adapting to emerging

threats dynamically. The results show that InstaGuard

achieved an overall accuracy of 95%, significantly

outperforming traditional methods that typically

range between 70% and 85%. Specifically, fake

account detection using Random Forest and XGBoost

reached an accuracy of 92.1%, while scam detection

through BERT and RoBERTa achieved 93.4%.

Phishing link detection using LSTM and URL

reputation-based models recorded the highest

accuracy of 95.2%. Impersonation detection,

implemented using CNN and Siamese Networks,

attained an accuracy of 94.5%, and misinformation

classification with graph-based NLP and RoBERTa

reached 95%. These results demonstrate the

robustness of InstaGuard in accurately identifying

fraudulent activities with minimal false positives.

5.2 Comparative Analysis with Existing

Approaches

InstaGuard was benchmarked against traditional rule-

based fraud detection systems and deep learning-

based fraud classifiers to evaluate its advantages.

Conventional fraud detection systems primarily rely

on static datasets, making them ineffective against

evolving fraud patterns. Rule-based systems, which

use predefined heuristics such as engagement

thresholds or keyword filters, are limited in their

ability to detect new scam techniques. Traditional

supervised learning models, including Logistic

Regression and Decision Trees, have been applied to

fraud detection but often achieve around 75%

accuracy due to feature dependencies, dataset

constraints, limited adaptability to evolving

fraudulent patterns, and challenges in handling

complex data distributions

5.3 Impact of Real-Time Data

Collection

A major advantage of InstaGuard is its exclusive use

of real-time data through Instagram’s Graph API,

which significantly enhances fraud detection.

Traditional fraud detection models rely on static

datasets, which become obsolete over time, leading to

reduced accuracy in identifying emerging fraud

techniques. By continuously collecting data in real-

time, InstaGuard ensures that the models remain

updated and responsive to the latest fraudulent

activities. Additionally, real-time behavioral analysis

improves fraud detection precision, reducing false

positives by distinguishing genuine user interactions

from automated or deceptive engagements. The

system’s ability to retrain models dynamically

ensures that it adapts to evolving fraud tactics,

making it more resilient against sophisticated

fraudulent activities.

5.4 Fraud Score Analysis & Threshold

Optimization

To enhance detection reliability, InstaGuard assigns a

fraud probability score to each analyzed profile, post,

or link. The effectiveness of this scoring system was

evaluated through extensive threshold optimization

experiments. A lower fraud threshold of 50% resulted

in high recall (98%) but increased false positives,

leading to unnecessary flagging of legitimate

accounts. In contrast, a higher threshold of 90%

yielded high precision (99%) but lower recall,

missing some fraudulent cases. The optimal fraud

detection threshold was determined to be 75%, which

ensures a balanced trade-off between accuracy and

minimal misclassification. By refining this threshold,

InstaGuard minimizes false alarms while maintaining

high detection reliability, making it an effective tool

for identifying fraudulent content and profiles on

Instagram.

5.5 Limitations & Challenges

Despite its high accuracy, InstaGuard faces several

challenges. One of the primary concerns is data

privacy and ethical consideration One of the main

concerns is data privacy and ethical considerations.

Although the system strictly processes publicly

available data following Instagram's privacy policies,

ongoing concerns about data security, ethical

implications, potential misuse, user consent,

regulatory compliance, transparency, accountability,

and evolving legal frameworks still persists. While

the system processes only publicly available data in

compliance with Instagram’s privacy policies,

concerns about data security and ethical implications

remain. Future improvements will focus on

integrating differential privacy techniques to enhance

user data protection. Another challenge is the

possibility of fraudsters developing sophisticated

evasion tactics to bypass detection. To address this

InstaGuard will incorporate adversarial learning

techniques to enhance model resilience against

evolving threats. Additionally, real-time fraud

detection requires significant computational

resources, which may lead to scalability concerns.

Cloud-based model deployment and edge computing

InstaGuard: An AI-Powered Framework for Detecting Fraudulent Activities on Instagram Using Machine Learning and Deep Learning

Techniques

845

strategies will be explored to optimize computational

efficiency and ensure scalability for large-scale

implementation.

6 CONCLUSION AND FUTURE

WORKS:

6.1 Conclusion

InstaGuard is an AI-driven fraud detection system for

Instagram, using machine learning and deep learning

to identify fake accounts, scams, phishing,

impersonation, and misinformation. It leverages real-

time data for accurate and adaptive fraud detection.

Future improvements include cross-platform

detection, privacy-preserving techniques, and

enhanced scalability the system effectively identifies

various forms of fraud, including fake accounts,

scams, phishing attempts, impersonation, and

misinformation. By utilizing real-time data collection

through Instagram’s Graph API, InstaGuard ensures

up-to-date fraud detection, reducing the limitations of

static datasets. The multi-model approach, which

includes Random Forest,XGBoost, BERT,

RoBERTa, LSTM, CNN, and Siamese Networks,

enhances detection accuracy across different fraud

types.

The experimental results demonstrate the

robustness of InstaGuard, achieving an overall

accuracy of 95%, significantly improving fraud

detection efficiency. The fraud probability scoring

mechanism further refines detection by providing a

dynamic and risk-based evaluation of user profiles,

content, and interactions. Compared to existing fraud

detection methods, InstaGuard’s real-time analysis

and adaptive learning capabilities provide superior

accuracy and adaptability to evolving fraud tactics.

Despite these achievements, challenges such as

data privacy concerns, adversarial evasion tactics,

and computational scalability remain. Addressing

these challenges through advanced privacy-

preserving techniques and improved fraud response

mechanisms will further enhance InstaGuard’s

effectiveness.

6.2 Future Works

To further improve InstaGuard and expand its

applications, several future enhancements are

planned.

Integration of Graph Neural Networks (GNNs):

Future versions of InstaGuard will incorporate GNNs

to analyze account relationships and identify

coordinated fraudulent activities. This will enhance

the detection of bot networks and fraud rings

operating on Instagram.

Adversarial Learning for Fraud Evasion Handling:

Fraudsters continuously adapt their tactics to evade

detection. InstaGuard will implement adversarial

learning techniques to train models against evolving

fraudulent behaviors, making the system more

resilient.

Cross-Platform Fraud Detection: By adapting to

various platforms, this method ensures extensive

coverage, enhancing fraud detection effectiveness

and security across diverse digital environments.

Privacy-Preserving AI Techniques: To enhance user

data protection, InstaGuard will integrate differential

privacy and federated learning approaches. This will

allow fraud detection without compromising user

privacy, aligning with ethical AI principles.

Scalability and Cloud-Based Deployment:

InstaGuard’s computational efficiency will be

improved through cloud-based deployment and edge

computing strategies, ensuring scalability for large-

scale fraud detection.

User Feedback and Model Improvement: Future

versions of InstaGuard will incorporate a feedback

mechanism that allows users to report

misclassifications. This feedback loop will

continuously refine the models and improve detection

accuracy.

By implementing these enhancements,

InstaGuard will evolve into a more robust, scalable,

and adaptable fraud detection framework. The

continuous integration of advanced AI techniques and

real-time monitoring will strengthen its ability to

combat fraudulent activities effectively. With its high

accuracy and adaptability, InstaGuard has the

potential to become a standard fraud detection system

for social media platforms, ensuring a safer digital

space for users worldwide.

REFERENCES

Al-Qurishi, M., Alrubaian, M., Al-Amri, J., & Alamri, A.

(2018). Sybil Defense Techniques in Online Social

Networks: A Survey. IEEE Access, 6, 12079-12094.

Chu, Z., Gianvecchio, S., Wang, H., & Jajodia, S. (2012).

Detecting Automation of Twitter Accounts: Are You a

Human, Bot, or Cyborg? IEEE Transactions on

Dependable and Secure Computing, 9(6), 811-824.

Cresci, S., Di Pietro, R., Petrocchi, M., Spognardi, A., &

Tesconi, M. (2017). The paradigm-shift of social spam:

From traditional spam to fake news. Future Generation

Computer Systems, 90, 46-58.

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

846

Cui, L., Zhang, D., Wang, D., & Lee, D. (2019).

Fakenewsnet: A data repository with news content,

social context, and dynamic information for fake news

detection. Big Data, 7(1), 67-79.

Jain, M., Kaur, G., & Kapoor, R. (2022). Cybersecurity in

Social Media: Emerging Trends and Challenges.

Proceedings of the International Conference on Cyber

Threat Intelligence, 122-135.

Kaghazgaran, P., Caverlee, J., & Alfifi, M. (2018).

Combating fake followers in social media through self-

disclosure analysis. Proceedings of the 41st

International ACM SIGIR Conference on Research &

Development in Information Retrieval, 210-219.

Kumar, S., Spezzano, F., Subrahmanian, V. S., &

Faloutsos, C. (2016). Edge weight prediction in

weighted signed networks. IEEE Transactions on

Knowledge and Data Engineering, 29(1), 44-57.

Lee, S., & Kim, J. (2014). WarningBird: Detecting

Suspicious URLs in Twitter Stream. IEEE Transactions

on Dependable and Secure Computing, 11(3), 280-293.

Liu, Y., Wu, Y. F., & Meng, X. (2017). Understanding

misinformation diffusion on social media: The effects

of message and network characteristics. Proceedings of

the ACM Conference on Computer-Supported

Cooperative Work, 931-943.

R. N. V. Jagan Mohan, B. H. V. S. Rama Krishnam Raju,

V. Chandra Sekhar, T. V. K. P Prasad. "Algorithms in

Advanced Artificial Intelligence - Proceedings of

International Conference on Algorithms in Advanced

Artificial Intelligence (ICAAAI-2024)",

CRC Press, 2025.

Rath, B., Maharana, A., Mohapatra, S., & Panda, G. (2019).

Detection of Fake Accounts in Social Networks using

Machine Learning Algorithms. IEEE International

Conference on Information Technology, 23-28.

Varol, O., Ferrara, E., Davis, C., Menczer, F., & Flammini,

A. (2017). Online Human-Bot Interactions: Detection,

Estimation, and Characterization. Proceedings of the

Eleventh International AAAI Conference on Web and

Social Media, 280-289.

Zhang, X., Chao, L., Wang, Z., & Li, H. (2021). Detecting

Malicious URLs in Social Media Based on LSTM

Networks and Attention Mechanisms. IEEE Access, 9,

20120-20132.

Zhou, X., & Zafarani, R. (2019). Fake news detection: A

data mining perspective. ACM Computing Surveys

(CSUR), 51(4), 1-41.

InstaGuard: An AI-Powered Framework for Detecting Fraudulent Activities on Instagram Using Machine Learning and Deep Learning

Techniques

847