Alzheimer’s disease (AD) is a degenerative neurological condition characterized by cognitive decline, memory loss, and reduced everyday function, which eventually causes dementia. Symptoms develop years after the disease begins, making early detection difficult. While AD remains incurable, timely detection and prompt treatment can substantially slow its progression. This study presented a framework for automated AD detection using brain MRIs. Firstly, the deep network information (i.e., features) were extracted using various deep-learning networks. The information extracted from the best deep networks (EfficientNet-b0 and MobileNet-v2) were merged using the canonical correlation approach (CCA). The CCA-based fused features resulted in an enhanced classification performance of 94.7% with a large feature vector size (i.e., 2532). To remove the redundant features from the CCA-based fused feature vector, the binary-enhanced WOA was utilized for optimal feature selection, which yielded an average accuracy of 98.12 ± 0.52 (mean ± standard deviation) with only 953 features. The results were compared with other optimal feature selection techniques, showing that the binary-enhanced WOA results are statistically significant (p < 0.01). The ablation study was also performed to show the significance of each step of the proposed methodology. Furthermore, the comparison shows the superiority and high classification performance of the proposed automated AD detection approach, suggesting that the hybrid approach may help doctors with dementia detection and staging.

Oral cancer, also known as oral squamous cell carcinoma (OSCC), is one of the most prevalent types of cancer and caused 177,757 deaths worldwide in 2020, as reported by the World Health Organization. Early detection and identification of OSCC are highly correlated with survival rates. Therefore, this study presents an automatic image-processing-based machine learning approach for OSCC detection. Histopathological images were used to compute deep features using various pretrained models. Based on the classification performance, the best features (ResNet-101 and EfficientNet-b0) were merged using the canonical correlation feature fusion approach, resulting in an enhanced classification performance. Additionally, the binary-improved Haris Hawks optimization (b-IHHO) algorithm was used to eliminate redundant features and further enhance the classification performance, leading to a high classification rate of 97.78% for OSCC. The b-IHHO trained the k-nearest neighbors model with an average feature vector size of only 899. A comparison with other wrapper-based feature selection approaches showed that the b-IHHO results were statistically more stable, reliable, and significant (p < 0.01). Moreover, comparisons with those other state-of-the-art (SOTA) approaches indicated that the b-IHHO model offered better results, suggesting that the proposed framework may be applicable in clinical settings to aid doctors in OSCC detection.

When dealing with big data, “divide and conquer” is the most commonly used strategy in practice to partition a big dataset into such smaller subsets that each subset can be handled by a computer or a node of cluster or cloud computing systems. However, among many existing partitioning or sampling techniques, it is not clear which one is suitable and how the size of subset may affect the performance of further analysis. In this paper, after presenting a generic framework of ensemble approach for learning from big data, we focus our investigations on systematically evaluating the effect of partitioning strategies and subset size on ensemble performance. The experimental results have demonstrated that three investigated partitioning / sampling strategies behaved statistically similar but the subset size may affect the performance of the ensemble in very drastically different ways, which are grouped into three patterns, rather than just one default perception - the bigger the better.

Divide-and-Conquer is probably the most commonly used strategy to deal with a big data that is too big to be loaded into any computing system's memory as a whole for analysis. It partitions such a big dataset into many smaller subsets that can be loaded into computer memory separately to induce models, which can be combined by machine learning ensemble methods. However, it is not clear that how the size of subsets may affect the learning performance of individual models and their ensemble. This paper proposes an ensemble based algorithm to quickly detect their relational patterns in terms of ensemble accuracy and the size of partitioned data subset. An ensemble framework of the algorithm is implemented and tested on 12 relatively big benchmark datasets. The experimental results indicate that it is able to identify the relation patterns accurately and efficiently in less than 10 steps. The identified patterns show that in most cases it is not necessary to use the whole big dataset for analysis as few smaller subsets are already sufficiently representative of the underlying problem, which is obviously a useful knowledge in big data analysis.

Pilgrim safety during Hajj is challenged by dense crowds, sustained physical exertion, and demanding environmental conditions that can increase tiredness and compromise well-being. Although wearable sensing enables objective monitoring, the reviewed literature provides limited evidence on how synthetic augmentation affects predictive performance and cross-participant generalization when wearable data are scarce. To address this gap, this study presents an integrated framework that combines correlation- and domain-informed feature screening with statistically constrained synthetic data generation and a complementary assessment of predictive performance and synthetic data fidelity. Using the Hajj 1445 (June 2024) Apple Watch dataset comprising 120 records from three participants, the study compares classical resampling, univariate normal generation, multivariate normal (MVN) sampling, and covariance-aware generation using Cholesky decomposition and Mahalanobis distance filtering under a consistent classifier evaluation protocol. The highest observed record-level test accuracy is 93.33%, achieved by a Decision Tree using MVN sampling with moderate clipping ($\pm 1.4$). Leave-one-participant-out evaluation of the same configuration yields a mean accuracy of 68.65% (SD = 19.72%), indicating lower cross-participant generalization. The regularized MVN–Cholesky–Mahalanobis configuration attains a composite synthetic-data quality score of 79.27%. These results show that statistically constrained synthetic augmentation can support tiredness prediction when real data are limited, while the participant-grouped results highlight the need for validation using larger and more diverse Hajj cohorts.