http://hdl.handle.net/123456789/10315 Sat, 17 Oct 2026 03:42:09 GMT 2026-10-17T03:42:09Z http://hdl.handle.net/123456789/21343 Mental Health Chatbot With AI Muhammad Usman Iftikhar, 01-132202-032; Ahmad Faraz, 01-132212-004; Sadeed Ullah, 01-132212-038 An increasing number of individuals face mental health problems, including anxiety, stress, and depression because academic, social, and personal demands keep rising. The conventional mental health services encounter multiple obstacles such as restricted accessibility as well as cost barriers and societal stigma that limit people from obtaining timely assistance. This project developed an AI-enabled mental health chatbot mobile app that provides scalable, accessible, and interactive methods for mental health evaluation and support for users. This Mobile app presents the Depression Anxiety Stress Scales (DASS-21) questionnaire in a conversational format, allowing users to conduct daily self-assessments. User interaction with this chatbot runs through its empathetic transformer-based system, which determines user sentiment using sentiment analysis to produce weekly mental health indexes. The application stores generated insights in Firebase, enabling users to track their progress over time and identify patterns in their mental health. Additionally, the app offers customized coping strategies and resources based on individuals’ responses, ensuring that users receive targeted support that suits their specific needs. This personalized approach not only fosters a deeper understanding of one’s mental well-being but also empowers users to take proactive steps towards improvement. Users update their self-analyses, conducted in the mental health application on a regular basis, they get a better idea of their emotional patterns. These self-estimates records also will be available to appointed therapists providing more rational decisions and individual care under the control of a human. Users benefit from the app because it implements intelligent communication to detect mental health conditions while building self-awareness and increasing involvement with mental health services. Supervised by Engr. Muhammad Kashif Naseer Thu, 01 Jan 2026 00:00:00 GMT http://hdl.handle.net/123456789/21343 2026-01-01T00:00:00Z http://hdl.handle.net/123456789/21801 Eeg-Based Vr Smart Learning System Muhammad Abu Bakar Roshan, 01-132222-025; Syed Muhammad Abdullah, 01-132222-043; Zarrar Ali, 01-132222-050 Monitoring student engagement during a lecture objectively is diffcult. Instructors depend on visual cues that are easy to miss, self-report questionnaires that interrupt the session, and post-hoc performance data that arrives too late to act on. This thesis addresses that gap with an EEG-based engagement monitoring system built for virtual reality learning environments and evaluated under real cross-subject deployment conditions. EEG data were collected from 16 participants using an OpenBCI Cyton board (8 channels, 250 Hz) while they watched academic lectures through a Meta Quest 2 headset. Each participant completed six sessions across six undergraduate subjects at three diffculty levels, giving four cognitive conditions per session. Recordings were preprocessed with a 50 Hz notch flter, a 1–40 Hz band-pass flter, artifact rejection, and 2- second Hamming windows with 50% overlap. Pipeline A assigned four-class cognitive load labels, producing 80,640 windows. Pipeline B computed 24-dimensional Welch PSD feature vectors and assigned binary engaged/disengaged labels via the Pope Engagement Index. Three novel deep learning architectures were evaluated under a fve-fold stratifed subject-independent cross-validation protocol. EEG-HybridFormer, combining a multi-scale convolutional front-end with a Transformer encoder and DINO selfsupervised training, achieved the highest mean accuracy of 85.64%. EEG-Deformer++ reached 78.45% and EEG-DINO++ established a 50.66% unsupervised baseline. An ablation study confrmed that subject-independent evaluation reduces accuracy by 12– 15 percentage points relative to a random split. A group disengagement detection module flags timestamps where six or more participants are disengaged for six or more consecutive seconds. Hard sessions consistently produced more disengagement zones, peaking at 13 of 16 students in the most diffcult Digital Logic Design session. A real-time inference pipeline using BrainFlow, a FastAPI WebSocket server, and a React dashboard demonstrates live deployment. The results show that subject-independent EEG engagement classifcation is feasible from a low-channel consumer-grade device, and that content diffculty reliably predicts grouplevel disengagement in a VR lecture setting Supervised by Dr. Waleed Manzoor Thu, 01 Jan 2026 00:00:00 GMT http://hdl.handle.net/123456789/21801 2026-01-01T00:00:00Z http://hdl.handle.net/123456789/21797 Vr-Eeg System for Gamified Assessment Ayesha Yasir, 01-132222-010; Isha Naeem, 01-132222-019; Muhammad Umer, 01-132222-031 Assessing the cognitive workload is a fundamental need for the design of intelligent and adaptive human computer interaction (HCI) systems, especially in immersed environments where the difficulty level of a task can dynamically affect a user’s mental state. In this research, a VR-EEG framework for cognitive workload assessment based on the electroencephalography (EEG) signal during a memory-based VR task is introduced, which is gamified to make the task more realistic. The EEG data were recorded from 16 participants using an OpenBCI Cyton 8-channel biosensing board when participants were asked to complete tasks under four different workload conditions: Relaxation, Easy, Medium, and Hard. Three different deep learning models were designed and tested under the same experimental setup. The first model, MultiBranch Attentive EEGNet (MBA-EEGNet), uses multi-branch temporal feature extraction, depthwise spatial convolution and channel attention to efficiently learn discriminative EEG representations. The second model is Multi-Scale Dynamic Graph Cross-Attention Conformer (MS-DGCA Conformer), which combines multi-scale convolution, dynamic graph-based spatial learning, cross-attention, and Conformerbased feature extraction to model the spatial and temporal relationships of EEG signals. Last, the proposed MS-DGCA-MBA-EEGNet Hybrid model is an integration of complementary features of MBA-EEGNet and MS-DGCA Conformer that aims to enhance the representation of both temporal EEG dynamics and spatial connectivity patterns. The evaluation with the topics proved the mean classification accuracy of MBA-EEGNet as 79.23% and that of MS-DGCA Conformer as 76.74% when subject wise evaluation was performed. The MS-DGCA Conformer also showed a good mean accuracy of 78.28% across 5 folds of the cross validation, which shows that it is generalizing well. The proposed MS-DGCA-MBA-EEGNet Hybrid model performed best on all the subjects (81.50% mean, across the 16 participants) compared to the individual architectures. The findings in this work show the power of this new approach that integrates convolutional, attention-based, graph-based, and Conformer-based feature learning to provide accurate multi-level cognitive workload estimation in immersive VR applications. A cognitive workload assessment of VR users in the past required the researcher to observe from afar, carry VR equipment, or choose the appropriate participant. Keywords: EEG, Cognitive Workload Assessment, Virtual Reality, MBA-EEGNet, MS-DGCA Conformer, Hybrid Deep Learning, Brain–Computer Interface, EEG Classification, Multi-Scale Feature Learning Supervised by Dr. Waleed Manzoor Thu, 01 Jan 2026 00:00:00 GMT http://hdl.handle.net/123456789/21797 2026-01-01T00:00:00Z http://hdl.handle.net/123456789/21803 Smartrad-Xai Muhammad Waleed Asrar, 01-132212-051; Abdullah Aamir, 01-132222-002; Hameed Junior, 01-132222-012 The early detection of chest diseases from X-ray images vital for timely clinical decision making especially in resource-limited health care systems where the availability of expert radiologists can be limited. The current computer-aided diagnostic systems have some drawbacks, including limited real-time access, limited clinical validation and reduced generalization between diverse disease categories. This study presents an Automated Chest Disease Classification framework, named SmartRadXAI, which combines deep learning and the mobile health care technology to address the challenges and provide reliable and easy-to-use diagnosis. In this research work, we introduce the automated chest disease classification system based on pretrained deep learning models: ResNet18, ResNet50, Vision Transformer (ViT), YOLOv8s, and YOLOv8n, which is called SmartRad-XAI. The fiveclass disease classification was done on ChestX-ray14 dataset from the NIH. By enhance accuray and precision of model, the data was preprocessed using Roboflow, which involves resizing images and augmenting data sets. Standard classification metrics were used to train all models and compare their results. Experiments demonstrated that YOLOv8s had the best performance, and the light version YOLOv8n was not as accurate as the other models. Besides the deep learning framework, a mobile app was designed to offer real-time chest X-ray analysis. The app enables users to upload photos by using the camera or gallery, then get predictions of which diseases are detected, with levels of confidence, and manage their prediction history. In addition, the patients X-Ray results were verified with the help of Islamabad Diagnostic Centre (IDC) where its predictions were compared with the interpretation of the patient by expert radiologist for the purpose of assessing its applicability in practical applications. As a whole, the study SmartRad-XAI highlights the potential of merging deep learning with mobile healthcare technology to deliver reliable, efficient, and accessible AI-powered chest disease diagnosis both in experimental and real clinical settings Supervised by Dr. Amna Waheed Thu, 01 Jan 2026 00:00:00 GMT http://hdl.handle.net/123456789/21803 2026-01-01T00:00:00Z