Journal of Computers, Mechanical and Management

Advances In Adaptive Machine Learning Algorithms for Enhanced Security In IoT Networks

Jayashri Jayesh Patil, Ramkumar Solanki — Sat, 28 Feb 2026 00:00:00 +0300

The accelerated growth and diversification of Internet of Things (IoT) environments have compounded security issues that are caused by non-stationary traffic, dynamic attacker tactics, and extreme resource constraints. In this case, the failure of the existing process for preventing intrusion through conventional, manually configured intrusion detection systems is becoming less sustainable, while adaptive machine learning (ML)-based security systems are becoming more popular. Nonetheless, the survey literature is generally more biased toward algorithmic enumeration or single-detection accuracy and provides very little critical evaluation of adaptation mechanisms, dataset realism, real-time capability, and implementation constraints. This paper provides a systematic, deployment-conscious review of adaptive machine learning methods for IoT security, conducted through a PRISMA-directed, systematic literature review. An integrative taxonomy of learning paradigms and adaptation methods is presented, a critical review of popular datasets on IoT security is conducted, and a discussion of performance metrics beyond accuracy, such as false positives, robustness, latency, and energy overhead, is provided. The review also examines real-time deployment issues related to edge-cloud, resource constraints, retraining costs, and orchestration complexity, as well as the security of adaptive models against adversarial manipulation and privacy leakage. The literature review, by reconceptualizing the problem of IoT intrusion detection as an adversarially exposed, adaptive learning task with deployment constraints, identifies gaps in the field and provides directions for future work to build a scalable, reliable, and dependable IoT security system.

A Systematic Review of Privacy-Aware Cloud Framework for Medical Secure E-Governance Data Processing

Qing Guan, Mustafa Muwafak Alobaedy, Mohd Nurul Hafiz Bin Ibrahim, S. B. Goyal — Sat, 28 Feb 2026 00:00:00 +0300

Cloud computing has greatly increased the effectiveness of e-governance, but there are also significant concerns about data security and privacy. The paper presents an in-depth evaluation of privacy-focused cloud architectures for the secure processing of medical e-governance data, in line with PRISMA. The study examined 72 peer-reviewed articles published after 2013 from IEEE Xplore, the ACM Digital Library, ScienceDirect, and SpringerLink. The study researched technologies, including AI-driven anomaly detection, hybrid cloud architecture, blockchain-enabled access management, and homomorphic encryption. This review organizes the available frameworks and evaluates how well they performed in previous studies. To build greater trust in digital governance systems, future trends point to lightweight encryption, cross-device functionality, and AI-powered security solutions. This in-depth examination of privacy-conscious frameworks identifies weaknesses in the research and offers helpful tips for both researchers and policymakers. The results indicate gaps in existing methodologies, thereby facilitating the development of e-governance infrastructures that are more secure, cost-efficient, and scalable, thereby enabling effective healthcare applications.

Structural Synthesis of Epicyclic Gear Trains by Deep Learning and Generative AI for Adaptive Automation

Sat, 28 Feb 2026 00:00:00 +0300

Structural synthesis of Epicyclic Gear Trains (EGTs) is a computationally demanding activity, particularly when identifying isomorphism between complex topologies and producing new gear designs to support high-performance automation systems. Graph-theoretic and algebraic methods are traditional and involve manual intervention and duplicate solutions. To address this shortcoming, this paper proposes a DL-based Generative AI framework for the automated synthesis and classification of EGTs. A Generative Adversarial Network (GAN) is trained on existing EGT topologies, learning their structures, creating new feasible structural mechanisms, and identifying duplication through degree sequence estimation and graph matching. The strategy is combined with the calculation of the connectivity matrix and the representation of the structural graph to ensure manufacturability and kinematic feasibility. The effectiveness of the proposed AI model is validated by the analysis of different EGTs, with 4--5 links, single DOF. Findings demonstrate that the GAN-based synthesis reliably distinguishes structurally distinct gear trains, eliminates pseudo-isomorphic designs, and saves a significant amount of design time. The technique justifies adaptive automation by designing intelligent mechanisms that require minimal human intervention. The paper demonstrates that AI-based synthesis can be highly effective in next-generation smart factories, robotic actuation, transmission systems, and reconfigurable automation platforms.

Brand Loyalty Drivers among Generation Z Fashion Consumers

Sat, 28 Feb 2026 00:00:00 +0300

Brand loyalty is crucial in the competitive fashion market, particularly among Generation Z. Although previous studies have investigated what drives loyalty, there is still limited evidence from India, particularly about gender differences. This study adopts a context-specific and exploratory approach to examine brand loyalty and its drivers among Generation Z fashion consumers in Bangalore. The study adopts a quantitative research design with a structured questionnaire using a 5-point Likert scale. A sample of 100 Generation Z students in Bangalore was selected using convenience sampling to collect the data. Further descriptive and inferential statistical analyses were conducted using SPSS. The findings show positive associations among brand loyalty, brand awareness, perceived quality, emotional connection, and social influence. Independent-samples t-tests reveal no significant difference in overall brand loyalty between male and female respondents. However, regression analyses indicate that perceived quality and brand awareness are relatively stronger predictors of brand loyalty among male respondents. In contrast, emotional connection is a stronger predictor among female respondents. These findings suggest differences in motivational pathways rather than loyalty intensity. The study suggests that while overall brand loyalty levels are similar across genders, the motivational drivers underlying loyalty differ. These findings are context-specific and exploratory, and their generalizability is limited by convenience sampling and a restricted geographic scope.

Hierarchical Deep Learning Ensemble Framework for Multi-Class Rice Foliar Disease Diagnosis

Sat, 28 Feb 2026 00:00:00 +0300

Infestations of foliar diseases in rice plants are common and can reduce harvest yields and affect food supplies worldwide. A system for the automatic detection of these diseases was developed in this study using seven different deep learning models. Six common types of rice leaf diseases were tested using models such as EfficientNet (B0 and B7), ResNet50, InceptionV3, VGG16, and VGG19. The proposed framework integrates the advantages of all models, assigning greater significance to those that exhibit superior performance. By achieving 96.97% accuracy while retaining speed and lightweight features, MobileNetV2 demonstrated superior performance. Both InceptionV3 and EfficientNetB7 performed well, reporting accuracies of 96.78% and 96.40%, respectively. It was also observed that newer, more efficient models exhibited markedly superior performance compared to older deep networks. This method makes it easier to bridge the gap between the urgent need for rapid disease detection on farms and the lack of agricultural experience. The system, which uses low-cost equipment, helps small farmers all over the world diagnose diseases accurately, resulting in better yields of crops.

Advancing Brain Tumor Detection: Optimized Machine Learning Models for Enhanced Diagnostic Accuracy

Sat, 28 Feb 2026 00:00:00 +0300

Segmentation of brain tumors from MRI continues to be difficult because tumors are different and there are not enough of each type. This implementation study improves Mask R-CNN for BraTS2020 by using three new ideas: a ResNet101 backbone that was trained on RSNA pneumonia data (Adam lr=0.001, batch=2 on RTX 3060), MRI-specific augmentation (57,195 2D slices from 369 3D volumes), and one-class loss weighting (λmask = 2.0) tuned to a 9:1 background tumor-pixel ratio to manage the imbalance in BraTS2020. This approach improved recall by 6 points compared with λmask = 1.0. With λmask = 1.0, the recall value is 0.66, and for λmask = 2.0, the recall value is increased to 0.72. Five-fold cross-validation shows that the results are stable (Dice=0.75, p < 0.01 vs ImageNet baseline), with performance by region: core=0.72, edema=0.68, and enhancing=0.76, and probability calibration characterized by an Expected Calibration Error (ECE) of 0.82 under a coarse, three-bin reliability analysis. To balance high-sensitivity tumor detection with a recall of 0.72 on the BraTS2020 dataset, our proposed method considered resource constraints for real-time deployment at 15 FPS.

Editorial Comments

Ritesh Bhat — Sat, 28 Feb 2026 00:00:00 +0300