IJRC https://ir.kdu.ac.lk/handle/345/8908 2026-04-08T13:41:34Z 2026-04-08T13:41:34Z Kumari, HMLS Kumari, HMNS Nawarathne, UMMPK https://ir.kdu.ac.lk/handle/345/8915 2026-02-19T08:12:54Z 2025-07-01T00:00:00Z

Speech Emotion Recognition with Hybrid CNNLSTM and Transformers Models: Evaluating the Hybrid Model Using Grad-CAM Kumari, HMLS; Kumari, HMNS; Nawarathne, UMMPK Emotional recognition and classification using artificial intelligence (AI) techniques play a crucial role in human-computer interaction (HCI). It enables the prediction of human emotions from audio signals with broad applications in psychology, medicine, education, entertainment, etc. This research focused on speech-emotion recognition (SER) by employing classification methods and transformer models using the Toronto Emotional Speech Set (TESS). Initially, acoustic features were extracted using different feature extraction techniques, including chroma, Mel-scaled spectrogram, contrast features, and Mel Frequency Cepstral Coefficients (MFCCs) from the audio dataset. Then, this study employed a Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and a hybrid CNN-LSTM model to classify emotions. To compare the performance of these models, classical image transformer models such as ViT (Visual Image Transformer) and BEiT (Bidirectional Encoder Representation of Images) were employed on the Mel-spectograms derived from the same dataset. Evaluation metrics such as accuracy, precision, recall, and F1-score were calculated for each of these models to ensure a comprehensive performance comparison. According to the results, the hybrid model performed better than other models by achieving an accuracy of 99.01%, while the CNN, LSTM, ViT, and BEiT models demonstrated accuracies of 95.37%, 98.57%, 98%, and 98.3%, respectively. To interpret the output of this hybrid model and to provide visual explanations of its predictions, the Grad-CAM (Gradient-weighted Class Activation Mappings) was obtained. This technique reduced the black-box character of deep models, making them more reliable to use in clinical and other delicate contexts. In conclusion, the hybrid CNN-LSTM model showed strong performance in audio-based emotion classification.

2025-07-01T00:00:00Z Luxshi, K https://ir.kdu.ac.lk/handle/345/8914 2026-02-19T08:15:07Z 2025-07-01T00:00:00Z

AI-Driven Disaster Prediction and Early Warning Systems: A Systematic Literature Review Luxshi, K Numerous advancements in artificial intelligence drive better accuracy and improved performance of disaster prediction as well as early warning systems for hazards. This review collects and integrates contemporary findings regarding AI management of disasters through machine learning along with deep learning along with data analytics techniques which address natural disasters and human-made emergencies. The paper analyzes how artificial intelligence contributes to earthquake forecasting processes while also providing information regarding flood forecasting and wildfire detection systems and other hazard assessment needs. This research studies how AI technology links with Internet of Things (IoT) and remote sensing systems for conducting real-time disaster surveillance. The discussion includes thorough assessments of important barriers which include issues with data quality together with system limitations and moral concerns. Future researchers can use this study to determine ways that will enhance AI-based disaster resilience strategies.

2025-07-01T00:00:00Z Kumari, HMLS https://ir.kdu.ac.lk/handle/345/8913 2026-02-19T08:15:32Z 2025-07-01T00:00:00Z

Innovative ECG Classification Approach Utilizing a Transfer Learning-Driven Ensemble Architecture Kumari, HMLS An electrocardiogram (ECG/EKG) is a vital methodology that is used for the diagnosis and monitoring of heart diseases by recording the electrical activity of the heart. However, manual analysis of ECGs shows limitations such as noise sensitivity, visual interpretation constraints and data imbalance. The proposed study a deep learning ensemble model combining DenseNet121, InceptionV3, and ResNet50 are implement to classify ECG images to improve diagnostic accuracy. The model is trained on two datasets: the National Heart Foundation 2023 ECG dataset and the ECG Dataset for Heart Condition Classification, focusing the main cardiac conditions such as abnormal heartbeat, myocardial infarction. The preprocessing techniques include background removal of ECG signal images, grayscale conversion, and data augmentation to enhance image quality and overfitting reduction. Stratified 5-Fold cross-validation was employed to demonstrate the generalization abilities of the proposed models. Early stopping and performance plots demonstrated that proposed model is not overfitting and two proposed models show consistent accuracy which suggests the model is not biased toward a specific dataset. While the ensemble models, as demonstrated in this study, produce better results than single models. The proposed study demonstrates validation accuracies of 98.62% and 96.75% for the National Heart Foundation 2023 dataset and the ECG dataset for heart condition classification, respectively, using 5-fold stratified crossvalidation. There are still some limitations, such as the proposed ensemble models not being evaluated using Explainable AI, which reduces clinical trust. Additionally, small datasets can limit the model's generalizability. Therefore, this study demonstrates the potential of deep ensemble models with advanced preprocessing for ECG classification, but it also highlights the importance of greater transparency, better dataset diversity, and real-world validation in future research studies.

2025-07-01T00:00:00Z Luxshi, K Rathnayaka, RMKT Seneviratna, DMKN Kithulwatta, WMCJT https://ir.kdu.ac.lk/handle/345/8912 2026-02-19T08:14:17Z 2025-07-01T00:00:00Z

Artificial Intelligence in Smart Cities and Urban Mobility: A Systematic Literature Review Luxshi, K; Rathnayaka, RMKT; Seneviratna, DMKN; Kithulwatta, WMCJT Artificial intelligence (AI) has been pivotal in advancing urban mobility and smart city planning. It offers innovative solutions to address emerging challenges in urban areas. With the global metropolitan population expected to comprise approximately 70% by 2050, the need for efficient, sustainable, and accessible urban mobility systems has become increasingly urgent. This systematic review synthesized 50 peer-reviewed studies from 2015 to 2024 that explore the implementation of AI alongside Internet-of-Things and Information Communication Technology in urban mobility. In particular, it highlights research on real-time traffic signal optimization, predictive algorithms, and intelligent routing systems, which have proven effective in reducing traffic congestion, improving the efficiency of public transportation, and enhancing safety through self-driving vehicles. Key challenges in implementing AI within smart cities and urban mobility include concerns over data privacy and sharing, infrastructure inadequacies, and the digital divide between regions. This systematic review has identified to overcome these obstacles, future research should focus on exploring innovative AI pathways, ensuring equitable access to AI technologies, and strengthening the physical infrastructure necessary to support smart city initiatives worldwide.

2025-07-01T00:00:00Z