http://hdl.handle.net/123456789/13168 2026-04-04T12:27:48Z 2026-04-04T12:27:48Z Ali Arshad, 01-243231-002 http://hdl.handle.net/123456789/19737 2025-10-08T03:57:48Z 2025-01-01T00:00:00Z

Exploring Evolutionary Algorithms for Optimal Features Selection to Detect Anomaly Based Intrusion In IoT Ali Arshad, 01-243231-002 In light of the fast-evolving scenario of the IoT, network security constantly bears immense challenges due to the increasing number of cyber threats and vulnerabilities. Traditional intrusion detection systems use predefined signatures and rule-based approaches to detect malicious activities within a network. In contrast to the traditional signature-based intrusion detection system, which utilizes previously established attack patterns, an anomaly-based intrusion detection system typically utilizes machine learning, statistical models, and artificial intelligence to assess network traffic, system log entries, and user behavior. . This research proposes the anomaly-based intrusion detection system using Particle Swarm Optimization (PSO)-based feature selection and ensemble learning in enhancing detection accuracy for IoT networks. Performance of stacking, hard voting, soft voting, and autoencoder-based models is evaluated over the benchmark datasets NSL-KDD and KDDCup 99 in analyzing their effectiveness in detecting anomalous behaviors in IoT environments. From the results, it is clear that PSO-based feature selection is highly significant in anomaly detection. Anomaly detection gets better performance by reducing feature redundancy along with improving classification accuracy. Of all the models tested, Stacking performed the best, with an accuracy of 98.87% on NSL-KDD and 99.76% on KDDCup 99, proving to be the most effective method. Soft Voting and Hard Voting also did well on NSL-KDD, recording 98.38% and 98.13% accuracy respectively, which highlighted the strength of ensemble methods in identifying anomalies based on IoT. The Autoencoder model demonstrated unsupervised anomaly detection, with a slightly less accurate of 96.63% on NSL-KDD and 98.23% on KDDCup 99, due to its greater false positive rates. This suggests that models based on deep learning could make a significant performance improvement on anomaly classification by leveraging their explicit feature selection capabilities. Stacking is an ensemble learning algorithm that increases predictive performance by aggregating multiple base models in a superior way. It can be used optimally when there exists some labeled data in supervised learning cases. Auto-Encoders correspondingly are neural networks deployed mainly in unsupervised learning, and they serve very well in detecting anomalies without labeled data. Supervised by Dr. Saba Mahmood

2025-01-01T00:00:00Z Hafiz Muhammad Ahmad, 01-243222-003 http://hdl.handle.net/123456789/19738 2025-10-07T11:57:44Z 2025-01-01T00:00:00Z

AI-Driven for Smart Concrete Bridge Inspection and Maintenance Hafiz Muhammad Ahmad, 01-243222-003 Concrete bridges require timely and accurate defect detection in order to maintain their structural safety, time-consuming, expensive processes often suffer from human errors. Meta-learning is applied to enhance convolutional neural networks for multi-target defect classification. The CODEBRIM dataset is harnessed in this study, challenging due to overlapping defects and varying environmental conditions. MetaQNN and ENAS are two advanced neural architecture search techniques used to automatically design optimized CNN models. These models are then compared with traditional architectures like ResNet, VGG, and DenseNet. Experimental results show that meta-learned models achieve significant improvements in terms of classification accuracy and computational efficiency. The most accurate models attained 75% test accuracy, reflecting the strength of simultaneous multi-defect identification. It proposes an AI-powered way of automated bridge inspection that is expected to be faster, more reliable, and less expensive than today, enhancing security on infrastructure items and lower maintenance costs. Supervised by Dr. Usman Hashmi

2025-01-01T00:00:00Z Muhammad Idrees, 01-243222-007 http://hdl.handle.net/123456789/19739 2025-10-08T06:59:50Z 2025-01-01T00:00:00Z

Intelligent Resource Management Framework Using Context Aware Statistics for B5G Network Muhammad Idrees, 01-243222-007 With the rapid evolution of mobile communication from 1G to 6G, efficient resource management has become essential to meet increasing demands of the networks. Traditional static allocation methods struggle to handle dynamic and heterogeneous B5G networks, leading to inefficiencies in latency, bandwidth utilization, and Quality of Service (QoS). To address these challenges, we propose an Intelligent Resource Management Framework that integrates Federated Learning (FL) with Context-Aware Statistics for optimized resource allocation. Our approach enables the training of decentralized models at the network edge using the Federated Average (FedAvg) algorithm, reducing communication overhead while preserving data privacy. Each network node locally train a model on real-time context-aware data, including user mobility, traffic variations, and signal fluctuations. The local models are then aggregated to improve learning without compromising privacy. The experimental results demonstrate superior performance, achieving 99.58% precision in resource prediction, significantly outperforming traditional centralized deep learning models. Key performance matrices such as precision, recall, F1 score, and AUC confirm the model’s ability to efficiently allocate resources under varying network conditions. The framework dynamically adjusts to traffic surges, interference, and mobility changes, ensuring optimal QoS and network stability. FD averaging algorithm performs better than other techniques. So, the proposed work achieved resource management, privacy preservation, low latency, and high data rates. Supervised by Dr. Muhammad Khurram Ehsan

2025-01-01T00:00:00Z Muhammad Hashim, 01-243222-006 http://hdl.handle.net/123456789/19741 2025-10-08T05:22:09Z 2025-01-01T00:00:00Z

Precision Agriculture: Harnessing ML/DL for Rice Leaf Disease Diagnosis Muhammad Hashim, 01-243222-006 Machine Learning (ML) and Deep Learning (DL), more advanced techniques, are being studied in this particular paper as a method for early recognition and classification of leaf diseases on rice plants only for four common types of leaf diseases, such as Brown Spot, Hispa, Leaf Blast, and Bacterial Blight. The purpose was to devise an efficient and automated approach to detecting these situational diseases in Pakistan. This paper utilizes the YOLO11s model with geometrical transformations and other data augmentations for hyperparameter optimization. For the collection of objects under investigation, such as brown spots and other leaf diseases, such as the leaves of rice, a set of about six thousand (6000) pictures was collected. When these images were grown from those categories, this is made up of four disease types. Each set of images demanded pre-processing such as flipping, rotation, noise addition, HSV adjustments, and mosaic augmentation so that when presented under different conditions, the model is not only memorized but also generalized. The YOLO11s model was intentionally designed to be fast while avoiding diminishing accuracy as too many layers are added, which also brings about complexity and slow computation. New strategies like MixUp, Copy-Paste, and geometric operations were employed to make YOLO11s yield better results. The tasks above had a comprehensive model training a staggering 400 epochs, stopped early at 331, and the epoch 281 model was picked up as the best. The model evaluation phase, on the other hand, featured quantitative evaluation measures such as Mean Average Precision (mAP), precision, recall, f1 score, and runtime efficiency ratios over all stages of the model. The other method folds to 87.7% mAP@0.5, which has been the case for the best of the other tested equations CNNoll, YOLOv5, YOLOv8 Rice. The model in question had a precision of 0.859 and a recall of 0.795 with an inference speed of 3.1 ms per frame of images, making it possible to apply live agricultural systems. Compared to conventional methods like SVM, KNN, Decision Trees, and Random Forests, the YOLO11s model provides superior accuracy and significantly faster inference speed, making it more practical for real-world implementation. This research contributes to precision agriculture by offering a reliable, efficient tool for rice disease detection, helping farmers manage crop diseases more effectively, thereby increasing yield and reducing economic losses. Future work could involve integrating IoT for continuous monitoring, expanding the dataset for broader disease coverage, and developing lightweight model versions for deployment on mobile and edge devices in resource-limited agricultural environments. Supervised by Dr. Usman Hashmi

2025-01-01T00:00:00Z