UMT Artificial Intelligence Review

Towards Robust and Explainable Early Rumor Detection: A Noise-Aware Graph Learning Framework

ABU BAKAR SHABBIR, Usama Husnain — Mon, 20 Apr 2026 00:00:00 +0500

Social media platforms have changed the communication process in such a radical way that now it takes just a few moments for unverified or fake news to reach millions of users. While in the past couple of years various methods have been developed to fight against the spread of false rumors, there are still many challenges to face. These challenges include dealing with noisy data, fast detection, and offering actionable insights for giving quick solutions. The current study aimed to put forward a conceptual noise-aware graph-based framework for the early detection of false rumors on social media. The framework melds together robustness, propagation-aware graph representations, and lightweight early-warning mechanisms in a single design. In order to depict the propagation patterns, the framework makes use of Graph Neural Networks (GNNs) for graph-based propagation modeling and also supports an uncertainty estimation module to handle data biases and incompleteness. One of the main features of the framework is the pinpointing of the most influential posts, users, and propagation paths that have a direct effect on the detection outcomes. While this work is not centered around experimental results, it lays out in detail the real-world validation plan using benchmark datasets, such as Twitter15/16, Weibo, and FakeNewsNet. The integration of noise-awareness, early detection, and interpretability in a single framework is the first step in the direction of robust and explainable rumor detection, thus opening new avenues for future theoretical and experimental work in this field.

Hybrid Feature-based Machine Learning Framework for Automated Brain Tumor Classification

Anam Naveed — Wed, 13 May 2026 12:18:20 +0500

The classification of brain tumors using Magnetic Resonance Imaging (MRI) is crucial for early diagnosis and efficient decision-making in clinical practice. However, manual analysis is a time-consuming process that may introduce observer bias. To address these issues, the current study presented a hybrid feature-based Machine Learning (ML) model for automated brain tumor classification, integrating both handcrafted and Deep Learning (DL) features to enhance robustness and accuracy. The suggested method combines Histogram of Oriented Gradients (HOG) to retrieve local texture data and high-level semantic details obtained with the help of the ResNet-50 deep convolutional neural network (CNN). The hybrid feature vectors are then classified with various ML classifiers, such as Linear, Gaussian, and Quadratic Support Vector Machines (SVMs), Logistic Regression (LR), Random Forest (RF), and XGBoost. The model was tested using two publicly-available benchmark datasets, viz., the Figshare Brain Tumor MRI dataset and Harvard Brain Tumor MRI dataset of a total of 10,286 MRI images. The experimental findings showed that the hybrid framework is best in the Figshare dataset, where Quadratic SVM has the best classification accuracy of 97%, and the other models are 96% and 95% with Gaussian SVM and LR, respectively. RF yields optimal accuracy on the more difficult Harvard dataset at 76%, which means that the proposed approach can be generalized to diverse data distributions. A further study is an ablation study, which supports the claim that the combination of handcrafted and deep features is a significant performance metric on classification when compared to the features separately. The findings confirmed that the hybrid framework has been developed as a suitable, robust, and universalized framework to classify automated multi-class brain tumors with the help of MRI images.

Streamlining Recruitment with AI: Design and Implementation of a Web Based Smart Hiring System

Wed, 13 May 2026 15:39:08 +0500

With the technological advancement, the traditional recruitment processes have become considerably dependable on manual labour and prone to bias, which is quite time consuming and challenging. Since the competition is getting tough with each passing day and the new talent is emerging, it is necessary to address such problems and find a solution that may automate such work with precision. The current study aimed to discuss HireForge: A Smart Hiring System that presents a web-based AI-assisted solution. This is specifically designed to aid and streamline the job-seeking and recruitment processes by smartly analysing the HireForge resumes and matching the candidate’s data according to the job openings and their requirements. Job seekers submit their professional information through a structured form, eliminating the requirement of file uploads. The system stores this information in the Firestore Database, and the AI logic then evaluates the job seeker’s information against the specific requirements of the job. This provides a clean and quick list of shortlisted resumes along with the details of score matching (0-100%) against the job specification. After the selection of accepted resumes, the candidate is evaluated by an AI-generated test according to the job role. It supports both the recruiters and the job seekers by providing in-depth details of skills, experiences, and role stability. This project also aimed to demonstrate the integration of AI with the web technologies. Moreover, the potential of smart systems was also highlighted to help and automate human resource (HR) processes as well as reduce recruitment biases.