Fake Reviews Detection on E-Commerce Websites Using Novel User Behavioral Features: An Experimental Study

Authors: N Mughal, G Mujtaba, MH Mughal, A Manaf, Z Kamangar. Published in ACM Transactions on Asian and Low-Resource Language Information Processing.

Abstract: The trend of writing fake reviews has recently increased with the rapid growth of e-commerce websites. Fake reviews are usually written to promote or demote the targeted products to affect the customer's decision and thus achieve a competitive advantage. Several techniques have been proposed to detect fake reviews written in English, and promising results have been obtained in the literature. Nevertheless, detecting fake reviews for low-resource languages (such as Roman Urdu) is still in the infancy stage and suffers from low classification results for two main reasons. Firstly, the existing studies mostly worked on textual features or lingual features. Secondly, the datasets used in existing studies are highly imbalanced, and proper attention to this issue may further enhance the performance. Therefore, to address these weaknesses and further enhance the performance, we have identified three types of discriminative features: review textual features, review lingual features, and review behavioral features using the Daraz dataset. Moreover, we evaluated LSTM-based text generation techniques for textual features and the random undersampling and oversampling for behavioral and lingual features to deal with class imbalance problems. Finally, we empirically evaluated the performance of machine learning and deep learning algorithms in classifying fake reviews written in the Roman Urdu language. The experimental results show that user behavioral features play a vital role in detecting fake reviews. Moreover, it was found that text generation is ineffective for balancing the textual data because the informative feature for fake review detection depends on the user behavioral features compared to textual features. Finally, the experimental results show that gradient boosting (GB) outperformed other models and improved 3% accuracy from the baseline study.

Aerial Image Classification in Post Flood Scenarios Using Robust Deep Learning and Explainable Artificial Intelligence

Authors: A Manaf, N Mughal, K R Talpur, B A Talpur, G Mujtaba, S R Talpur. Published in IEEE Access.

Abstract: Efficiently delivering timely assistance to flooded regions is a critical imperative, and leveraging deep-learning methodologies has demonstrated significant efficacy in addressing environmental challenges. Several authors have collected data from specific regions and presented the methodologies for classifying flooded images. However, there are two main limitations of existing methodologies and benchmark datasets. Firstly, the models are trained on the images collected from specific geographical regions, which limits their ability to generalize when encountering images with diverse features or from varied regions. Secondly, the models are trained on high-resolution images and lack classification for the low-resolution images. In this study, we curated a dataset by merging benchmark datasets and acquiring images from web repositories. The main objective of this study is to address resolution issues and enhance model accuracy across diverse regions. We conducted a comparative analysis of the curated dataset using various deep-learning models based on CNN architecture. The experimental findings revealed that MobileNet and Xception outperformed ResNet-50, VGG-16, and Inception(v3) models, achieving an impressive accuracy rate of approximately 98% and an f1-score of 92% for flood class. Additionally, we have also used XAI Lime to interpret model results.

Exploring the Impact of Self-Supervised Learning and Swin-B Transformer on Saliency Maps and Performance in Multi-Label Chest X-Ray Classification

Authors: A Manaf (co-author), AS Imran, G Mujtaba, et al. Published in IEEE. Research conducted at NTNU (NORPART-CONNECT) with the ISA Research Group.

Abstract: This study explores self-supervised learning (SSL) and Swin-B transformer architectures for multi-label chest X-ray (CXR) classification, with a focus on saliency maps and model interpretability. We compared Vision-Language Models and ROI-guided pipelines on medical imaging. The Swin-B SSL model achieved AUC 0.83 and IoU 0.16 on CXR pathology detection. An open-source web application was developed for clinician use.

Enhancing Wrist Abnormality Detection with YOLO: Analysis of State-of-the-Art Single-Stage Detection Models

Authors: A Ahmed, AS Imran, A Manaf, Z Kastrati, SM Daudpota. Published in Biomedical Signal Processing and Control 93, 106144.

Abstract: Diagnosing and treating abnormalities in the wrist, specifically distal radius, and ulna fractures, is a crucial concern among children, adolescents, and young adults, with a higher incidence rate during puberty. However, the scarcity of radiologists and the lack of specialized training among medical professionals pose a significant risk to patient care. This problem is further exacerbated by the rising number of imaging studies and limited access to specialist reporting in certain regions. This highlights the need for innovative solutions to improve the diagnosis and treatment of wrist abnormalities. Automated wrist fracture detection using object detection has shown potential, but current studies mainly use two-stage detection methods with limited evidence for single-stage effectiveness. This study employs state-of-the-art single-stage deep neural network-based detection models YOLOv5, YOLOv6, YOLOv7, and YOLOv8 to …

Small Data, Big Impact: A Multi-Locale Bone Fracture Detection on an Extremely Limited Dataset Via Crack-Informed YOLOv9 Variants

Authors: A Ahmed, AS Imran, A Manaf, Z Kastrati. Presented at the 21st International Conference on Frontiers of Information Technology (FIT 2024), Islamabad, Pakistan, December 9-10.

Abstract: Automated wrist fracture recognition has become a crucial research area due to the challenge of accurate X-ray interpretation in clinical settings. This study utilizes an extremely limited multi-region fracture dataset and proposes a novel approach where YOLOv9 is pre-trained on surface cracks rather than COCO. This method achieves state-of-the-art (SOTA) performance on the newly released FracAtlas dataset, improving the mean average precision (mAP) score by 7% and sensitivity by 13%.

Note: The full text of this paper is not yet publicly available.

AI-Enhanced Pediatric Pneumonia Detection: A CNN-Based Approach Using Data Augmentation and Generative Adversarial Networks (GANs)

Authors: A Manaf, N Mughal. Preprint available at arXiv:2507.09759.

Abstract: Pneumonia is a leading cause of mortality in children under five, requiring accurate chest X-ray diagnosis. This study presents a machine learning-based Pediatric Chest Pneumonia Classification System to assist healthcare professionals in diagnosing pneumonia from chest X-ray images. The CNN-based model was trained on 5,863 labeled chest X-ray images from children aged 0-5 years from the Guangzhou Women and Children's Medical Center. To address limited data, we applied augmentation techniques (rotation, zooming, shear, horizontal flipping) and employed GANs to generate synthetic images, addressing class imbalance. The system achieved optimal performance using combined original, augmented, and GAN-generated data, evaluated through accuracy and F1 score metrics. The final model was deployed via a Flask web application, enabling real-time classification with probability estimates. Results demonstrate the potential of deep learning and GANs in improving diagnostic accuracy and efficiency for pediatric pneumonia classification, particularly valuable in resource-limited clinical settings.

Pediatric Wrist Fracture Detection in X-rays via YOLOv10 Algorithm and Dual Label Assignment System

Authors: A Ahmed, A Manaf. Preprint available at arXiv:2407.15689.

Abstract: Wrist fractures are highly prevalent among children and can significantly impact their daily activities, such as attending school, participating in sports, and performing basic self-care tasks. If not treated properly, these fractures can result in chronic pain, reduced wrist functionality, and other long-term complications. Recently, advancements in object detection have shown promise in enhancing fracture detection, with systems achieving accuracy comparable to, or even surpassing, that of human radiologists. The YOLO series, in particular, has demonstrated notable success in this domain. This study is the first to provide a thorough evaluation of various YOLOv10 variants to assess their performance in detecting pediatric wrist fractures using the GRAZPEDWRI-DX dataset. It investigates how changes in model complexity, scaling the architecture, and implementing a dual-label assignment strategy can enhance detection performance. Experimental results indicate that our trained model achieved mean average precision (mAP@50-95) of 51.9%, surpassing the current YOLOv9 benchmark of 43.3% on this dataset. This represents an improvement of 8.6%.