Faculty of Computing Scopus 2

Permanent URI for this collectionhttps://rda.sliit.lk/handle/123456789/4896

Browse

Filters

20265

Advanced Search

Filter by

Settings

Start date: 2026End date: 2026

Search Results

Now showing 1 - 5 of 5
  • Thumbnail Image
    PublicationOpen Access
    Hybrid ABC–HBA feature optimization with self-training using simulated unlabelled data for robust intrusion detection
    (Elsevier Ltd, 2026) Harischandra, S; Rajapaksha, U.U. S; Silva, B.N; Jayawardena, C
    The increasing scale and heterogeneity of network traffic pose significant challenges for intrusion detection systems (IDS), particularly in detecting extremely rare attack classes and generalising to previously unseen threats under severe class imbalance. This study proposes a hybrid intrusion detection framework that integrates swarm intelligence–based feature optimisation with self-training using unlabelled data simulation to address these limitations. A novel ABC–HBA feature selection strategy is introduced, combining the efficient exploration capability of the Artificial Bee Colony (ABC) algorithm with the strong global exploitation and fast convergence of the Honey Badger Algorithm (HBA), resulting in a highly discriminative and compact feature subset. A Random Forest(RF) classifier augmented with a pseudo-labelling mechanism is then employed to enhance learning from unlabelled and unseen attack samples, enabling effective detection of novel attack patterns absent from the training set. To further mitigate extreme class imbalance, a hybrid resampling strategy is applied. Experimental evaluation on the KDD Cup 1999 dataset demonstrates that the proposed framework achieves an overall accuracy of 99.95% and a detection rate of 98.16%, while significantly improving the recognition of extremely rare attack classes, including a 92.86% detection rate for U2R attacks, which constitute less than 0.01% of the dataset. The proposed method consistently outperforms baseline RF, ABC-based, and several other state-of-the-art meta-heuristic and deep learning approaches, confirming its effectiveness in enhancing rare attack detection and generalisation to unseen threats in realistic intrusion detection scenarios.
  • Thumbnail Image
    PublicationOpen Access
    A physics-informed machine learning for detecting suspicious satellite maneuvers (orbital manipulation)
    (Elsevier B.V., 2026) Karunathilake K.K.H; Abeywardena, K.Y; Vecchini, S
    Satellite systems have become prime targets for cyberthreats given their critical role in global infrastructure and general lack of security. Among these, orbital manipulation, a form of satellite hijacking, is a particularly severe threat that can disrupt essential operations and impact national security. To address these concerns, this research proposes an Artificial Intelligence (AI)-based anomaly detection system that utilizes Machine Learning (ML) models to analyze telemetry data for possible orbital manipulations with a multi-gate physics architecture grounded in orbital mechanics, to verify that detected anomalies are kinematically inconsistent and are therefore genuine integrity failures. This research demonstrates that temporal-based models like LSTM are essential for this domain, achieving high recall rates which are then validated by the physics component. While the framework includes multiple physical constraints, this study specifically validates the energy-based Vis-Viva gate, with the Tsiolkovsky and Angular Momentum gates established as architectural designs for future verification. This study concludes that successful AI deployment in orbital cybersecurity requires a comprehensive approach that integrates domain-specific context and physics-informed validation beyond traditional performance metrics
  • Thumbnail Image
    PublicationEmbargo
    High-resolution optical imaging for sustainable fish freshness and safety assessment
    (Elsevier GmbH, 2026-04) Madhubhashini, M. N; Kahandawala, B.S; Sandaruwan, H.H.P. B; Silva, B.N; Wijenayake, U; Wijesinghe, R.E
    Fish freshness evaluation is crucial to ensure consumer safety, and rapid assessment is essential for effective and accurate quality control. To overcome the limitations of the gold standards, such as lack of structural depth information, high-time consumption, and labor-intensiveness, high-resolution Optical Coherence Tomography (OCT) was employed for real-time monitoring of fish freshness non-invasively. Microstructural changes of eye and skin of Indian Anchovies ( Stolephorus indicus ) specimens were considered as the main freshness parameters during refrigeration storage. Both eye and skin tissues exhibited decreased internal scattering, loss of clarity, boundary weakening, and gradual structural degradations through the OCT observations. The quantitatively assessed variance intensity, entropy, energy, and edge density clearly revealed the internal tissue disruption over storage time due to protein denaturation, oxidative damage, and fluid imbalance. The findings of this study indicate that OCT shows an insightful correlation with microbiological and biochemical spoilage processes, enabling the advanced identification of subtle microstructural changes in fish skin and eye, even at a prior stage of deterioration. Such capability offers an objective and rapid freshness evaluation approach that could greatly benefit supply chain management and post-harvest seafood quality monitoring.
  • Thumbnail Image
    PublicationOpen Access
    EuqAud: Detecting Gender Bias in Audio Datasets Using Polynomial Regression-Based Metric
    (Institute of Electrical and Electronics Engineers Inc., 2026) Jayawardena, S; Haddela, P.S; Shyamalee, T; Ekanayake, A; Mudalige, T; Dhanawardhana, I
    With the growing adoption of audio based AI systems in high-stakes domains such as healthcare, law enforcement, and social media, ensuring fairness particularly regarding gender bias has become critically important. While prior work on fairness has predominantly focused on disparities in model performance, bias inherent in training datasets remains underexplored. To address this gap, we propose EuqAud, a novel, pre-trained and traceable fairness metric that quantifies gender bias in audio datasets using raw acoustic features such as pitch, energy, amplitude, and voice activity. Unlike methods dependent on demographic labels such as race, age or language, EuqAud is designed to be demographic and language agnostic, enhancing its applicability across diverse contexts. The score is computed using an equation derived from polynomial regression with L2 regularization (Ridge regression), yielding robust and generalizable outputs. It spans a range from −10 to 10, where 0 denotes neutral, positive scores indicate male dominant bias, and negative scores reflect female dominant bias. For clarity, bias severity is categorized into three tiers: Neutral (EuqAud < 2), Moderate Bias (2 ≤ EuqAud ≤ 6), and Strong Bias (EuqAud > 6). Evaluation across multiple datasets demonstrates high predictive performance, with R2 values between 0.95 and 0.99. By focusing on dataset level bias rather than model outcomes, EuqAud offers a scalable and rigorous solution for advancing fairness in audio-based AI systems.
  • Thumbnail Image
    PublicationEmbargo
    Data-centric single teacher guided knowledge distillation for alleviating sub-optimal supervision in image classification
    (Elsevier Ltd, 2026-02-23) Sharma, K; Silva, B. N
    In recent years, larger, deeper, and more complex deep learning models have emerged as a result of advancements in deep learning techniques. Nevertheless, the computational costs have also increased with the growing model size. Thus, Knowledge Distillation has evolved into a cornerstone in contemporary machine learning, facilitating the transfer of knowledge from cumbersome teacher models to more compact student models. However, student learning is persistently challenged by sub-optimal supervision caused by erroneous and ambiguous teacher predictions. Moreover, the learning process is further deteriorated by the complications introduced through frequently encountered noisy labels in real-world datasets. Existing methods often resort to the ensemble of teachers, introducing additional complexity. We propose a novel, simple, and efficient learning method, Corrective Knowledge Distillation (CKD), to alleviate these drawbacks while relying solely on a single-teacher model. The proposed work employs a two-phase learning paradigm. In the initial phase, the teacher selectively teaches extremely confident knowledge to the student, and in the subsequent phase, the student leverages its own past learning experiences, conditioning its knowledge acquisition on the guidance of the teacher. The proposed method consistently exhibits superior performance in addressing sub-optimal supervision, as evidenced by comprehensive experiments on benchmark datasets such as CIFAR-100, CIFAR-100N-Fine, and ImageNet-1K. Notably, CKD surpasses established baselines, achieving substantial accuracy gains of up to 3.53% in real-world scenarios. Furthermore, CKD exhibits exceptional robustness in highly noisy environments, outperforming ensemble techniques by a significant margin of up to 5.18%. Our code is available at https://github.com/Karthick47v2/ckd.