Journal of Artificial Intelligence and Metaheuristics
Journal DOI
2833-5597ISSN (Online)
The ability to accurately classify retinal fundus images has been made possible by rapid improvements in deep learning (DL) and artificial intelligence (AI). This motivation led to developing a new AI-driven hybrid Convolutional Neural Network with Long Short-Term Memory (CNN-LSTM) architecture for precisely categorizing retinal diseases. The model first receives high-resolution retinal fundus images to extract various spatial properties, which are then processed by two parallel CNN branches after a standard convolutional layer. These branches use residual learning with convolutional and identity blocks to extract features. Following the reshaping and concatenation of the features from both branches, an LSTM layer captures inter-feature relationships. Eight retinal disorders are then predicted to belong to the same disease class via a fully linked classifier. Extensive simulations were run on a benchmark retinal OCT dataset, and performance was assessed using various criteria. The experimental results showed that the suggested hybrid model was adequate, with a high overall accuracy of 93% with F1-score values of 0.93, 0.94, and 0.93 for precision, recall, and accuracy, respectively. The model demonstrated considerable predictive abilities for all classes while perfectly classifying AMD, CNV, CSR, DME, DR, MH, and routine diseases to reveal its clinical value as an automated retinal processor.
Read MoreDoi: https://doi.org/10.54216/JAIM.100201
Vol. 10 Issue. 2 PP. 01-19, (2025)
Hierarchical clustering is applied in this research to study world COVID-19 data up to January 2025 and partition the primary clusters of countries based on epidemiological criteria. Total cases, deaths, recoveries, active cases, tests, population, and per-million were the data explored and were standardized and thereafter analyzed employing agglomerative hierarchical clustering with Ward linkage. The assessment yielded an average Silhouette of 38.5%, Davies–Bouldin value of 0.87, and Calinski–Harabasz value of 77.6, reflecting cluster validity in separation. The application of dendrograms and PCA projections to plot identified four clusters, reflecting differences in the severity of COVID-19 impacts and responses. Clustering analysis revealed that the high-burden clusters accounted for almost 45% of global death, while low-burden clusters were predominant in over 40% of nations with fewer than 100,000 accumulated instances. The outcomes illustrate hierarchical clustering as an unsupervised learning approach to analyzing epidemiological data and give quantitative estimates to facilitate comparative public health interventions across communities.
Read MoreDoi: https://doi.org/10.54216/JAIM.100202
Vol. 10 Issue. 2 PP. 20-31, (2025)
Accurate prediction of CO2 emissions from vehicles is essential for environmental regulation and sustainable transport design. Existing models often suffer from limited accuracy due to suboptimal hyperparameter configurations. This s tudy a ims t o e nhance C O2 e mission f orecasting b y c ombining d eep l earning with advanced metaheuristic optimization. An attention-based Encoder LSTM (EALSTM) model is trained on Canadian vehicle emissions data, with hyperparameters tuned using a novel Football Optimization Algorithm (FbOA), inspired by cooperative team dynamics in football. Comparative evaluation against eight other optimizers shows that FbOA achieves the best performance. The optimized EALSTM model yields an RMSE of 0.00349, MAE of 0.00010, and R2 of 0.984, outperforming all alternatives. These results demonstrate the effectiveness of domain-inspired metaheuristics in improving prediction accuracy. The proposed FbOA-EALSTM framework offers a scalable, accurate solution for emissions modeling and supports data-driven environmental policy and intelligent vehicle technologies.
Read MoreDoi: https://doi.org/10.54216/JAIM.100203
Vol. 10 Issue. 2 PP. 32-51, (2025)
The classification of brain tumors is crucial in the context of early intervention, as the appropriate and timely diagnosis can significantly influence the treatment plan and patient outcomes. Radiologists have long relied on their own judgment and have read these medical images through their own eyes, which is often subjective, time-consuming, and inter-observer variability is also likely to occur. Applications built on artificial intelligence (AI), or more specifically, deep learning (DL)-based algorithms, have radically changed the medical imaging field over the last couple of years and could potentially be used to automate the diagnosis process, offering prompt, trustworthy, and unbiased assessments. Despite such developments, most existing systems that rely on AI are constrained, especially when it comes to classification accuracy and robustness across different datasets. To overcome these problems, the article in this chapter presents a more effective DL model with a specifically designed architecture that aims to improve the classification of brain tumors. The specified methodology is based on preprocessing and data normalization steps that reduce noise and level out the data intensity, enabling effective feature extraction from the MRI images. This will increase the accuracy of the later classification. The primary component of the proposed methodology is an adapted version of DenseNet-201, designed explicitly for the four class brain tumor classification. To achieve optimal performance, the conventional output layer of DenseNet-201 was replaced with a Global Average Pooling (GAP) layer, designed to address the issues of vanishing gradients and overfitting commonly encountered during the training of deep networks. The architectural adjustment helps to combine the features and increase the overall generalization capacity of the model. The model was thoroughly tested using two datasets: one publicly available dataset on Figshare and a locally available dataset comprising a total of 3,504 T1-weighted contrast-enhanced MRI (T1-w MRI) images. The results of the experiment provided the proposed model with a general accuracy of 100 percent, which was higher than that of the existing comparative methods. Such results support the idea that complex architectural adjustments with the broader preprocessing strategy can be effective, and why deep neural networks can be viewed as trustworthy diagnostic tools in clinical neuro-oncology, potentially achieving extremely high accuracy.
Read MoreDoi: https://doi.org/10.54216/JAIM.100204
Vol. 10 Issue. 2 PP. 52–66, (2025)
Integration of quantum-inspired algorithms in machine learning has opened up new horizons for improving predictive performance, efficiency, and scalability across a broad spectrum of application domains. This paper presents a comparative investigation between traditional machine learning techniques and quantum-inspired models. Experimental experiments demonstrate that quantum-inspired approaches exhibit higher accuracy, training effectiveness, and stability on difficult datasets than traditional methods. Results point towards higher convergence rates, shorter runtime, and enhanced generalization capacity in quantum-inspired models, realized in the form of enhanced accuracy, precision, recall, and F1-scores. Receiver operating characteristic (ROC) and precision–recall analyses further confirm the superior discriminative power of quantum-inspired approaches. Results point toward the potential of quantum-inspired machine learning as an interface between conventional algorithms and the new frontier of quantum computing with a stepping stone to future-proof intelligent systems.
Read MoreDoi: https://doi.org/10.54216/JAIM.100205
Vol. 10 Issue. 2 PP. 67-81, (2025)