http://repo.lib.sab.ac.lk:8080/xmlui/handle/susl/3775 Fostering Multidisciplinary Research and Innovation for a Sustainable Future 2026-04-17T06:01:48Z http://repo.lib.sab.ac.lk:8080/xmlui/handle/susl/4080 ICAPS Front Page 2023-05-30T00:00:00Z http://repo.lib.sab.ac.lk:8080/xmlui/handle/susl/4079 A Novel Intelligent Video Surveillance Mechanism to Real-Time Identify Abnormal Activities Chandrasekara, P.G.I.M.; Chathuranga, L.L.Gihan; Chathurangi, K.A.A.; Seneviratna, D.M.K.N.; Rathnayaka, R.M.K.T. The main reason for the existence of most anti-corruption laws today is the inability to address the root causes. Abnormal behaviors occur through robbery, corruption, murder, threats, etc. Proper solutions to these are implemented only after abnormal incidents occur. Some CCTV cameras support object detection, but nothing beyond that. Manual monitoring of CCTV footage for abnormal events is laborious and time-consuming. Therefore, this study aimed to develop a new method for real-time identification of abnormal behavior in fighting scenes using a 3D Convolutional Neural Network (CNN) based spatiotemporal autoencoder. Initially, the study suggested an intelligent video surveillance system which uses deep learning techniques, including facial expression detection with CNN and YOLO v7. However, the accuracy of facial expression detection alone is limited in the real world. The proposed video surveillance system accurately detects abnormal fights by comparing a specially prepared video stream to frames generated by an autoencoder. A model was created using TensorFlow and other libraries to identify fighting scenes in a video stream through spatio-temporal encoders. After studying the proposed method using three case studies respectively, the last case study was able to reach the desired result. They were also tested on three different publicly available datasets: fer2013.csv facial expression dataset, emotion-facial-expression dataset in the Roboflow library, and CUHK Avenue dataset. The three case studies aimed to detect abnormal behavior in real-time, and the last method proposed achieved a 72.56% accuracy in identifying fighting scenes. Furthermore, future research could be carried out on this approach by studying areas with highly reported fighting incidents and developing new models specifically for those areas. The proposed system has the potential to detect abnormal activities in real-time, which can be useful in addressing the problem of abnormal behavior in both public and private environments. 2023-05-30T00:00:00Z http://repo.lib.sab.ac.lk:8080/xmlui/handle/susl/4078 Machine Learning Approach for Football Match Results Prediction Kaluarachchi, K.N.; Premachandra, K.P.; Dissanayake, R.B.N. Precisely predicting sports results is a widely known challenge in the sports industry. It has now become the trend to predict individual sports as well as less predictable team sports such as football, volleyball, basketball etc. Predicting the outcome of a football match is an expanded area of research simply for the commercial assets involved in the betting process. Conventionally, the final outcome of a match was predicted by the field experts. However, today this approach is empowered by the growing amount of diverse football-related information that needs to be processed. In this study, we use various machine learning (ML) techniques to compare the prediction results of the German Bundesliga which is one of the most popular European Leagues. This study mainly discusses the comparison between the performances of different machine learning models used in previous studies. The data used in this study were collected from season 2008/2009 to season 2022/2023 of the German Bundesliga. In order to increase the accuracy of the models, new attributes were introduced by calculating the rolling averages of the previous matches. Logistic Regression, Decision Tree, Random Forest, Support Vector Machine, k-Nearest Neighbor, Gradient Boosting, and Na¨ıve Bayes are the ML techniques used to predict the results by partitioning the dataset into training and testing. Training dataset includes data from season 2008/2009 to 2017/2018 (66.67%) and testing dataset includes data from season 2018/2019 to 2022/2023 (33.33%). By using several evaluation metrics such as accuracy, precision, sensitivity, F-1 score, and mean squared error, the best performing model is chosen to make the predictions. The results show that Random Forest gives the maximum accuracy of 0.6146 with precision, and sensitivity of 0.5221, and 0.8495 respectively. It can be concluded that, introducing new features, Random Forest is the best method that can be used in match result prediction. 2023-05-30T00:00:00Z http://repo.lib.sab.ac.lk:8080/xmlui/handle/susl/4077 Improving Protein Secondary Structure Prediction Using an Ensemble of Recurrent and Convolutional Neural Networks with Evolutionary Profiles Mufassirin, M.M.M. Protein secondary structure prediction is a critical sub problem in computational biology and bioinformatics. The prediction of protein secondary structure has been extensively studied using various computational methods, including empirical and physics-based approaches and machine learning algorithms. With the advancement of deep learning methods, protein secondary structure classification accuracy has been substantially enhanced. Protein secondary structures are broadly classified into either 3-state (Q3) or 8- state (Q8) classes. This study proposes an approach that combines Convolutional Neural Network (CNN), bidirectional Long-Short-Term-Memory (BILSTM), and evolutionary protein profile input features to improve secondary structure prediction accuracy. The proposed model was trained and validated using the DNSS2 dataset and tested on three independent test datasets, CB513, CASP11, and CASP12. The performance of the model was compared with five state-of-the-art approaches, and the impact of combining different input features on the model’s performance was also evaluated. The proposed approach outperformed the state-of-the-art approaches, particularly for Q3 secondary structure prediction using PSSM, HMM, and 7PCP as input features. The ensemble of CNN and BILSTM achieved the highest Q3 score of 85.35% and Q8 score of 75.51% on the test set. The approach presented in this study combines deep neural networks with optimized hyper-parameters and protein evolutionary profile features to improve secondary structure prediction accuracy, which is a novel contribution to the field. The proposed model significantly improved the accuracy of protein secondary structure prediction compared to five state-of-the-art methods. The approach can be useful in various fields, including drug discovery, protein engineering, and functional annotation. 2023-05-30T00:00:00Z