International Research Journal of Multidisciplinary Technovation

Fuzzy Adaptive Bobcat Optimization Algorithm (FABOA) Feature Selection Model and Double Deep Q Network (DDQN) Learning for Big Data Analytics

Vijetha M, Maria Priscilla G — Wed, 21 Oct 2026 00:00:00 +0000

Big data analytics is challenged by high dimensionality, redundant attributes, heterogeneous data distributions, and the computational burden of processing large-scale datasets. The current filter, wrapper and matrix factorization approaches typically ignore the relationships among features or converge too early, are not very adaptive, and are not very efficient in searching. To fill this space, this study offers an Intelligent Feature Selection and Large-Scale Data Analytics Framework based on Fuzzy Adaptive Bobcat Optimization Algorithm–Double Deep Q-Network (FABOA–DDQN). In addition to the Bobcat Optimization Algorithm, FABOA introduces triangular fuzzy membership functions to adjust exploration and exploitation based on the diversity of the population, improvement in fitness and the number of iterations. Its wrapper objective takes into account predictive performance, feature-subset compactness and feature redundancy. DDQN is not just a sequential combination, but rather a policy-learning controller: Candidate feature subsets are states, adding or removing features or changing their values is an action, and the reward value is a measure of fitness improvement and feature reduction. The online and target networks alleviate Q-value overestimation and stabilize search decisions. The experimental results were obtained by applying a 70:15:15 training/validation/testing split of a widely used SVM evaluator, 30 independent trials and tested against NNMF, S4NMF, RLNMFAG, DBMD variants, BOA, GA, PSO, WOA, DQN, and DDQN. On Covertype, FABOA–DDQN achieved 98.34% precision, 96.77% recall, 97.55% F1-score, and 97.10% RI. It surpassed DBMD-DLM-BFGS in all four of these measures by as much as 1.95%, 3.44%, 2.69%, and 3.29%, respectively, across all four datasets while bringing the ratio of the selected features down to 31%.

Analysis of Gene Co-Expression Networks using Centrality Measures in order to Identify Significant Genes Correlated with Cancer Diagnosis

Seema Sandeep Redekar, Smita Dange, Madhura Vyawahare — Fri, 30 Oct 2026 00:00:00 +0000

One of the most aggressive primary brain tumors is glioblastoma multiforme (GBM), which has poor prognosis, and with limited therapeutic options. Identifying reliable molecular markers will be crucial to enhancing the prognostic value for GBM and to elucidating the biological mechanisms of GBM progression. While co-expression network analysis has been used extensively for biomarker discovery, traditional centrality measures in graph theory assign only topological measures to genes, and exclude any biological information based on patient survival. In this study, we suggest a biologically informed node-weighted centrality approach for incorporating survival-related biological information with the conventional network topology to enhance biomarker prioritization. This dataset contains DNA methylation profiles from the The Cancer Genome Atlas (TCGA) GBM cohort which was preprocessed to bring out informative probes followed by univariate Cox proportional hazards analysis to extract the genes associated with survival. To incorporate biological weights that are derived from survival, node-weighted versions of degree, closeness, betweenness and eigenvector centrality were formulated on the biological network. The proposed framework was then used to prioritize genes within the co-expression network that have a biological relevance. A total of 156 genes were identified in the analysis that are associated with survival, and ten candidate prognostic biomarkers (SEMA3B, MARK2, APS, PITPNM2, SFRP1, SLC44A2, PRLH, KRTAP4.2, CTSZ and DIP2C) were selected. These biomarkers were uniformly associated with overall survival and are biologically meaningful molecular signatures with survival data and network topology. This study introduces a biologically informed node-weighted centrality framework which integrates the biologically derived information from the survival analysis into the graph-theoretic network analysis for biomarker prioritization. The proposed methodology is an extension of our previously reported biomarker discovery pipeline, where the node weighting is introduced using a survival-guided approach, while the previously validated biological dataset and the biomarker set remain. This framework offers a computational approach to biomarker prioritization based on networks of biological relevance which can be extended to GBM and other complex diseases.

An Adaptive Multimodal Meta-Fusion Framework for Alzheimer's Disease and Frontotemporal Dementia Classification using EEG, MRI, and Cognitive Biomarkers

Lalita Wasudev Moharkar, Monika Bhagwat — Wed, 21 Oct 2026 00:00:00 +0000

A single-modality approach to neurodegenerative disease classification is often limited by heterogeneous clinical manifestations, modality-specific feature distributions, and incomplete data availability. This study proposes an Adaptive Multimodal Meta-Fusion (AMMF) framework integrating cognitive-demographic, structural MRI, and EEG biomarkers for classification of the Alzheimer's disease (AD), Frontotemporal Dementia (FTD), and Cognitively Normal (CN) participants. Three modality-specific cohorts from two public repositories were considered: cognitive-demographic data (n = 679), structural MRI data (n = 595), and EEG data (n = 194; BrainLat n = 106 and OpenNeuro n = 88). Cognitive-demographic features are prioritized using explainable XGBoost with SHAP-based feature analysis, MRI data are processed using ConvNeXt-Small for spatial representation learning, and EEG signals are analyzed using a Transformer-based architecture for temporal and functional representation learning. For the matched BrainLat cohort, the three modality branches generated an 18-dimensional cognitive-demographic representation, a 128-dimensional EEG representation, and a 768-dimensional MRI representation. The proposed framework then applied adaptive late fusion at the modality-probability level: each modality generated an independent three-class probability vector, and a modality-availability mask was used to renormalize the predefined modality weights of 0.20, 0.30, and 0.50 assigned to the cognitive-demographic, EEG, and MRI modalities, respectively. The resulting weighted class-probability vector was supplied to a Random Forest meta-classifier for final three-class classification. The complete multimodal evaluation was performed on 106 subject-level matched participants with all three modalities. The proposed framework achieved an accuracy of 99.06%, demonstrating the utility of combining complementary cognitive, functional, and structural biomarkers through adaptive probability-level meta-fusion for AD, FTD, and CN classification. A continuity-corrected McNemar test yielded χ² = 4.000 (p = 0.0455) for C4 versus C7 and χ² = 4.167 (p = 0.0412) for C5 versus C7, indicating statistically significant paired differences, while C6 versus C7 yielded χ² = 3.200 (p = 0.0736). Integration of cognitive-demographic, MRI, and EEG modalities for multimodal disease classification. The proposed framework integrates cognitive-demographic, MRI, and EEG modalities for multimodal disease classification by combining heterogeneous features in a unified representation space and applying Random Forest-based meta-classification to the fused modality probabilities.

Bendlet-Based Medical Image Fusion Using Modified Spatial Frequency and Contrast Visibility

Muralidhar Naick B, Leela Rani D, Vijaya Kishore V — Mon, 26 Oct 2026 00:00:00 +0000

Multimodal medical image fusion combines complementary anatomical and functional information from different imaging modalities into a single image to support clearer visualization and more informed clinical interpretation. Conventional fusion methods may introduce blurring, reduce contrast, or lose important edge and texture information, particularly when curved anatomical boundaries are not represented effectively. This study proposes a Bendlet-based medical image fusion framework that integrates the Bendlet Transform with Contrast Visibility (CV) and Modified Spatial Frequency (MSF) fusion rules. The source images are first decomposed into low- and high-frequency subbands using four multiscale levels and directional subbands of [8, 8, 16, 16]. A bending parameter of 0.5 is used to represent curved image structures, and the transformed coefficients are processed in non-overlapping 32 × 32 blocks. The low-frequency coefficients are fused using the CV rule to preserve intensity and contrast information, whereas the high-frequency coefficients are fused using the MSF criterion to retain edges, textures, and local structural details. The final fused image is reconstructed through the inverse Bendlet Transform. The proposed method is evaluated on five medical-image databases using qualitative inspection and objective fusion metrics, including entropy, standard deviation, mutual information, spatial frequency, fusion symmetry, structural similarity, and edge-based quality measures. The reported comparisons indicate that the proposed method preserves complementary information and produces visually clearer fused images than the comparison techniques considered in the study. These findings suggest that Bendlet-based decomposition combined with frequency-specific fusion rules is a promising approach for multimodal medical image fusion, although broader clinical validation and statistical analysis are required before diagnostic use.

IoT Assisted Multimodal Deep Learning Framework for Real-Time Onion Spoilage Detection

Pooja K, Reni A — Fri, 30 Oct 2026 00:00:00 +0000

Onion storage suffers from major challenges due to post-harvest losses (PHLs) in agricultural supply chains, especially in developing countries where continuous monitoring of surrounding conditions is limited. Standard storage monitoring systems are typically based on traditional static threshold rules (e.g., an excess of 60% relative humidity), and as such, do not have intelligent capabilities to make decisions. To assist with this gap in the current available options for storage monitoring, this study presents a multimodal Internet of Things (IoT) system combined with deep learning for real-time detection of onion spoilage. This unique approach uses environmental data from Internet of Things (IoT) sensors connected via an ESP32 microcontroller to measure temperature, humidity, moisture content, and gas concentration. It also takes images of the onions and uses Convolutional Neural Networks (CNNs) to classify the photos of the onions as they spoil. The application of Grad-CAM visualization of the features contributing to image classification will also assist with the interpretability of the developed models. The evaluation of the proposed framework employed a dataset of 2400 onion images categorized into four classifications: healthy, sprouted, spoiled and with black mold. Precision, recall and F1-score were 91.8%, 93.2% and 92.4%, respectively, for each of the four classifications using the proposed multimodal IoT and CNN framework. Overall, the system achieved an end-to-end latency of approximately 1.3 seconds; therefore, providing almost real-time monitoring of stored onions.

Flexible Cunning Search Optimization based Light Gradient Boost Machine Convolutional Neural Network for Harmonic Distortion Mitigation

Sushma V. Sangle, Prakash G. Burade — Mon, 26 Oct 2026 00:00:00 +0000

Traction networks are attractive due to their advantages such as step-down of high voltage, precise speed control, energy efficiency, and lower maintenance. However, the railway traction loads inject a massive amount of harmonics into the traction power system, which leads to impacts such as thermal stress, mechanical fatigue, capacitor failure, electromagnetic interference, and high energy consumption, degrading the system performance. Hence, Total Harmonic Distortion (THD) mitigation is mandatory to improve operational stability and reliability performance of the traction network. Therefore, the research work considers the Flexible Cunning Search optimization-based Light Gradient Boost Machine Convolutional Neural Network (FCS-optimized GBCN) model to mitigate the harmonic distortion in the Traction Power Supply System (TPSS). The developed model adopts the time-varying and non-linear nature of harmonics to perform a more accurate and robust harmonic mitigation task. Especially, the FCS-optimized GBCN model predicts the harmonic distortion signal to remove the harmonics, which makes a clean sinusoidal waveform. Furthermore, the Flexible Cunning Search (FCS) optimization improves the solution quality, reduces the probability, and finds optimal solutions, which contributes to enhancing the model performance. Consequently, the FCS-optimized GBCN Model outperforms the switching output, train 1 current, train 2 current, train 1 voltage, train 2 voltage, power, and THD values are 4997, 2388.64 A, 846.49 A, 472.97 V, 538.70 V, 28022 W, and 0.68, respectively.