International Journal for Applied Information Management

Optimization of Fraud Detection in E-Commerce: A CGAN Data Augmentation Approach to Address Class Imbalance

2025-06-15T15:56:04+00:00

The rapid growth of e-commerce has increased the risk of fraud in online transactions, resulting in significant financial losses and decreased consumer trust. One of the main challenges in fraud detection is data imbalance, where the number of legitimate transactions far exceeds fraudulent transactions. This imbalance causes machine learning models to fail in accurately identifying fraudulent transactions. This study aims to evaluate the effectiveness of Conditional Generative Adversarial Network (CGAN) in improving fraud detection performance in e-commerce through data augmentation. Two machine learning algorithms, Random Forest (RF) and XGBoost, were used to classify transactions in both the original imbalanced dataset and the dataset augmented with CGAN. The study uses key evaluation metrics, including accuracy, precision, recall, and F1-score, to measure the model's performance. The results show that data augmentation using CGAN significantly improved the performance of both models. RF on the augmented dataset achieved an accuracy of 99.96%, precision of 99.93%, recall of 99.99%, and F1-score of 99.96%. Meanwhile, XGBoost achieved an accuracy of 99.93%, precision of 99.91%, recall of 99.94%, and F1-score of 99.92%. The main contribution of this study is to demonstrate that CGAN can effectively address the challenge of data imbalance and improve the reliability of fraud detection systems in e-commerce. This approach has the potential to be applied in various sectors facing similar issues, such as anomaly detection in finance and cybersecurity.

Analysis of Demographic and Consumer Behavior Factors on Satisfaction with AI Technology Usage in Digital Retail Using the Random Forest Algorithm

2025-06-15T15:58:29+00:00

The rapid integration of artificial intelligence (AI) into digital retail has reshaped consumer interactions, enabling personalized services and operational enhancements. This study investigates the demographic and behavioral factors influencing consumer satisfaction with AI technologies in digital retail, using the Random Forest classification algorithm for predictive modeling. After comprehensive preprocessing and hyperparameter tuning through grid search cross-validation, the Random Forest model achieved an overall accuracy of 83%. While the model showed strong performance for predicting satisfied consumers yielding a precision of 0.84, recall of 0.97, and F1-score of 0.90, it performed poorly in identifying dissatisfied users, with a recall of only 0.27 and F1-score of 0.39, highlighting a class imbalance issue. Feature importance analysis revealed that experiential factors, particularly enhanced AI experience and preference for online services, significantly influenced satisfaction levels, whereas demographic variables such as age and gender had limited predictive value. These findings emphasize the need for digital retailers to focus on user-centric design and service personalization, rather than demographic segmentation alone, to enhance customer satisfaction and loyalty. Furthermore, the study contributes methodologically by demonstrating the effectiveness of Random Forest in handling complex consumer datasets and theoretically by validating TAM and Customer Satisfaction Theory in the context of AI adoption. Despite limitations related to class imbalance and sector-specific data, this research offers actionable insights for retailers, marketers, and system developers aiming to improve AI-driven service quality and consumer engagement. Future studies are encouraged to address these limitations through the inclusion of emotional and contextual variables and by expanding the analysis to other industries for broader applicability.

User Profiling Based on Financial Transaction Patterns: A Clustering Approach for User Segmentation

2025-06-15T15:59:43+00:00

User profiling based on financial transaction patterns is crucial for improving customer segmentation and personalizing financial services. This study uses clustering techniques, specifically K-means, to analyze transaction data and segment users based on transaction amounts, times, and types. Three clusters were identified, each demonstrating distinct transaction behaviors: Cluster 0, primarily focused on purchases and occurring early in the week; Cluster 1, which emphasizes transfers and higher transaction amounts, typically occurring mid-week; and Cluster 2, similar to Cluster 0 but with a preference for later-week transactions. The analysis demonstrates that transaction patterns, including amount, time, and type, provide valuable insights for targeting specific user groups with personalized marketing strategies and financial products. The study also highlights the importance of improving clustering accuracy, as indicated by the moderate Silhouette Score of 0.33, suggesting that further refinement in the clustering methodology could lead to more distinct user segments. The findings of this study emphasize the potential for clustering techniques to enhance user profiling, ultimately improving business strategies, customer satisfaction, and loyalty. Limitations of the study, including the dataset’s single-month duration, suggest that further research incorporating larger and more diverse datasets, as well as alternative clustering techniques, could offer deeper insights into user behavior and refine segmentation strategies.

Assessing Sentiment in YouTube Video Content: A Title and Description Analysis Approach to Analyze User Reactions

2025-06-15T16:00:13+00:00

This study investigates the relationship between sentiment in YouTube video titles and descriptions and user engagement metrics, such as view count, like count, and comment count. The findings reveal that videos with positive sentiment generally attract higher levels of engagement, including more views, likes, and comments, while videos with negative sentiment typically receive lower interaction levels. The research emphasizes the importance of emotionally resonant content, suggesting that content creators should focus on producing videos with positive emotional tones to maximize audience interaction. Additionally, the study highlights the significance of well-crafted titles and descriptions as key drivers of engagement, as these textual elements influence viewers' initial expectations and emotional reactions. However, the study is limited to analyzing titles and descriptions, which may not fully capture the emotional tone of the video itself. Future research should incorporate the actual video content and explore additional engagement metrics, such as shares and watch time, for a more comprehensive understanding of viewer behavior. Despite these limitations, the study provides valuable insights that can guide content creators in tailoring their video content and metadata to foster greater viewer engagement and content success.

Implementation of Machine Learning Algorithms for Detecting Bot and Fraudulent Accounts on Instagram Based on Public Profile Characteristics

2025-06-15T16:00:52+00:00

The rapid growth of Instagram as a social media platform has led to increased challenges related to fake accounts, including bots, spam, and scam profiles, which threaten the integrity and trustworthiness of online information. This study implements machine learning algorithms, particularly the Random Forest classifier, to detect and classify Instagram accounts into four categories: Real, Bot, Spam, and Scam, based on publicly available profile characteristics. A dataset of 15,000 Instagram profiles was collected and preprocessed, extracting features such as follower count, following count, posting frequency, and presence of profile information. The Random Forest model was trained and evaluated, achieving an accuracy of 97% with high precision and recall across all categories. Behavioral analysis revealed distinct patterns in following/follower ratios, posting activity, and mutual friends that differentiate genuine users from fake accounts. Feature importance ranking highlighted follower count as the most influential attribute for classification. The model demonstrated strong robustness through ROC and Precision-Recall curves, underscoring its effectiveness in a multiclass classification task. This approach not only enhances automated detection and moderation of malicious accounts but also contributes to maintaining a safer social media environment by mitigating misinformation and fraud. Future work could improve detection by incorporating temporal activity data, linguistic analysis, and real-time monitoring to adapt to evolving deceptive behaviors. Taken together, this study confirms the viability of machine learning methods in addressing the growing issue of fake accounts on Instagram, offering scalable and interpretable solutions for social media security.