International Research Journal of Multidisciplinary Technovation

Dual Cloud - Sturnus Optimized Intent-BERT with Randomized Secure Anonymization based Bibliographic Network Recommender for Citation Recommendation System

Nurjahan V.A — 2026-09-02

Citation recommendation (CR) systems for manuscripts employ automated systems to identify and suggest relevant scientific publications for effectively citing specific text passages; however, they face challenges related to the promotion of unreliable sources and the privacy of sensitive data. To address these limitations, a novel Dual Cloud Sturnus Optimized Intent Learning BERT with Randomized-Adversarial Secure Anonymization (SIL-BERT) based Bibliographic Network Recommender (BNR) is proposed in this study to enable secure citation recommendation. Among these issues, shilling attacks are particularly important because malicious actors exploit the mathematical foundations of collaborative filtering and manipulate the algorithms. To address this issue, a novel Sturnus Optimized Self-Intent BERT (SS-IBERT) is employed and it effectively reduces artificial inflation caused by algorithmic rhetoric and thereby mitigates the suppression of emerging researchers. Moreover, domain-specific interactions combined with re-identification through linkable attributes generate a distinctive chronological activity fingerprint, which allows anonymized citation logs to be cross-referenced with public metadata. Therefore, to deal with this, a Randomized-Style Adversarial Anonymization (R-SAA) is used, and it effectively suppresses the Small-N Behavioral Vulnerability (S-NBV) that creates elevated re-identification risks. The experimental results demonstrate that the proposed Dual Cloud SIL-BERT achieves a citation-recommendation accuracy of approximately 99%.

Experimental Investigation on the Engineering Properties of a Novel Geomaterial Containing Bottom Ash, Blast Furnace Slag and EPS Beads

Akhand Pratap Singh — 2026-09-15

This study presents the experimental findings on a Novel Geomaterial (NGM) developed using bottom ash as the primary constituent material. The NGM was made by mixing bottom ash with blast furnace slag and some expanded polystyrene beads, and it relied on ordinary Portland cement as the binder for real. Five different EPS bead contents 0.2%, 0.4%, 0.6%, 0.8% and 1.0% were considered as mix ratios. Blast furnace slag was incorporated at three replacement levels of 4%, 8%, and 12%. In addition, two cement to bottom ash weight ratios, 20% and 30%, were adopted in the study. The experimental program looked at compressive strength, the stress-strain behavior, density, plus the initial tangent modulus for the NGM while using various mix designs and different curing times. The researchers performed compressive strength tests on 70 mm cube samples that were cured for 7 days, 14 days, and 28 days. The results indicate that compressive strength consistently improved with increasing slag content, higher cement dosage, and longer curing duration across all mix ratios, with recorded strengths ranging from 135 kPa to 1020 kPa. The stress–strain behavior exhibited a nonlinear relationship, with compressive stress increasing progressively with axial strain. Furthermore, the density of the prepared NGM decreased from 993 kg/m³ to 611 kg/m³ depending on the mix ratios, while the initial tangent modulus showed a positive correlation with compressive strength. The findings show a kind of manageable trade-off between density and short-term compressive behavior, but the whole thing around environmental sustainability and field readiness is still up in the air, mostly because embodied carbon assessment, leachability checks, durability testing, and wider field-scale validation still need to be done.

Composition-Dependent Magnetic Reversal and Partial Exchange Coupling in SrFe12O19/FeCo Nanocomposites

Akshaya R — 2026-08-31

SrFe₁₂O₁₉/FeCo hard–soft magnetic nanocomposites with different hard-to-soft magnetic phase ratios were successfully synthesized using a combination of sol–gel autocombustion and wet-chemical methods, followed by physical mixing. X-ray diffraction analysis confirmed the formation and coexistence of the hexagonal SrFe₁₂O₁₉ hard magnetic phase and body-centred cubic FeCo soft magnetic phase, with no detectable crystalline impurity phases. The structural parameters showed a composition-dependent variation in crystallite size and lattice strain, indicating changes in crystallinity and structural distortion with FeCo incorporation. The magnetic properties exhibited a strong dependence on the hard–soft phase ratio. With increasing FeCo content, the coercivity decreased from 1795 to 866 Oe, which can be attributed to the increasing contribution of the magnetically soft FeCo phase and its easier magnetization reversal. The differential magnetization (dM/dH) curves displayed a dominant central switching peak accompanied by symmetric secondary features, suggesting the coexistence of different magnetization-reversal processes and supporting the presence of partial exchange coupling between the hard and soft magnetic phases. Furthermore, the effective magnetic anisotropy decreased with increasing FeCo content due to the reduced relative contribution of the high-anisotropy SrFe₁₂O₁₉ phase. Despite the decrease in coercivity, the maximum energy product increased from 0.09 to 0.18 MGOe, demonstrating that appropriate adjustment of the hard-to-soft phase ratio can improve the overall magnetic performance. These findings demonstrate that FeCo incorporation provides an effective approach for balancing saturation magnetization and coercivity, highlighting the potential of SrFe₁₂O₁₉/FeCo nanocomposites as rare-earth-free materials for permanent magnet applications.

Tuning the Charge-Transport Properties of CeTiO₃ Thin Films through Controllable Phase Transformations Induced by the JNSP Technique

Yogeshwaran A — 2026-09-11

CeTiO₃ thin films were synthesized by jet nebulizer spray pyrolysis (JNSP) at substrate temperatures of 350, 400, and 450 °C to investigate temperature-dependent structural, optical, morphological, and electrical properties for photodiode applications. X-ray diffraction analysis confirmed temperature-induced phase evolution and showed an increase in average crystallite size from 11.46 to 12.98 nm, accompanied by a reduction in dislocation density. FESEM observations revealed progressive grain growth and improved film densification, with average particle size increasing from approximately 17 to 35 nm. UV–Vis analysis demonstrated a decrease in optical band gap from 2.96 to 2.64 eV as the substrate temperature increased, indicating enhanced visible-light absorption. XPS confirmed the presence of Ce, Ti, and O with mixed Ce³⁺/Ce⁴⁺ states and predominantly Ti⁴⁺. CeTiO₃/p-Si photodiodes were fabricated and evaluated through current-voltage measurements under illumination. Device performance improved with increasing deposition temperature, with the 450 °C sample exhibiting the lowest ideality factor of 1.991, the highest barrier height of 0.411 eV, responsivity of 5.97 mA/W, quantum efficiency of 8.89%, and detectivity of 5.66 × 10^10 Jones. These results demonstrate that substrate-temperature control effectively tunes CeTiO₃ thin-film properties and enhances photodiode performance. The optimized films therefore show strong potential for oxide-based optoelectronic and photodetection device applications.

A Decomposition-Driven Deep Learning Model for Intelligent Irrigation Decision-Making

Rajesh Kumar — 2026-09-22

Traditional farming methods still used in many regions contribute to lower crop yields despite the availability of ample arable land. Integrating emerging technologies such as Machine Learning (ML) and the Internet of Things (IoT) into agriculture can enable intelligent irrigation and smart farming. These technologies support real-time monitoring of environmental parameters and help optimize irrigation schedules using soil moisture, temperature, and humidity data. IoT devices and ML algorithms can predict irrigation needs, reduce water wastage, and enhance crop yields. Therefore, this study proposes a hybrid model that combines a Convolutional Neural Network (CNN) with Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) for irrigation classification and water conservation in agricultural fields. CEEMDAN decomposes the input features into multiple frequency components, which are then fed to the CNN to improve learning performance. To evaluate the proposed approach, the study developed six models, including standalone and hybrid models. The proposed model achieved the highest accuracy (98%), precision (0.95), recall (1), and F1-score (0.97) compared with the benchmark models. Furthermore, the confusion matrix reveals minimal misclassification, while the ROC curve, with an area under the curve (AUC) of 0.995, confirms excellent discriminative capability. These findings demonstrate the effectiveness and robustness of the proposed model for improving irrigation decision-making and promoting sustainable water management in agriculture.

Deep Learning Methods for Multimodal Fake News Classification Combining Textual and Visual Information

Pundlik Dattatray Jadhav — 2026-09-03

The multimodal nature of fake news propagating in the social media has necessitated the need to have strong detection systems that can detect textual and visual discrepancies. In this work, the authors suggest a computationally efficient multimodal deep learning framework using Bidirectional Encoder Representations of Transformers (BERT) to extract textual features and convolutional neural networks (ResNet50, MobileNet and VGG16) to learn visual representations. It uses a feature-level fusion approach that involves the integration of contextual text embeddings and deep visual features, and a softmax-based classification layer. The Fakeddit dataset is experimented on a six-class classification configuration, with unequal data distribution. The suggested multimodal model (BERT + ResNet50) is more effective with an accuracy of 94.7% and a macro F1-score of 0.91, and recall, as compared to unimodal baselines. Image-only models are performing moderately (75-78% accuracy) and the text-only BERT model is at 88.3 percent accuracy which shows the significance of multimodal integration. The findings show that feature-level fusion is effective to capture cross-modal discrepancies, minimizing false positives and enhancing generalization. The presented framework offers a computational scaling alternative to attention-based models, which is computationally expensive, and forms a solid basis in future multimodal misinformation detection studies.

Reliable Drug Recommendations using a Hybrid CNN-BiLSTM and Whale Algorithm-Optimized Aspect-Based Sentiments

Suruchi Chawla — 2026-09-15

The growing availability of drug reviews from patients has paved the way for future patient-centric and data-enabled healthcare support; however, the majority of existing sentiment-based drug recommendation systems use overall sentiment (sentiment polarity) to generate drug recommendations. In this study, an aspect-aware and optimized deep-learning-based drug recommendation system is proposed using sentiment analysis. A combined Convolutional Neural Network and Bidirectional Long Short-Term Memory architecture is employed to leverage important local semantics as well as contextual and sequential dependencies from drug reviews. The model is extended into an explainable drug recommendation system that maps user-reported symptoms to patient-opinion-informed drug rankings. Sentiments are analyzed at the aspect level, focusing on effectiveness, side effects, dosage, safety, and cost. To achieve better results, the Whale Optimization Algorithm is used to assign optimal weights to each aspect, thereby ranking drugs according to data-driven aspect preferences learned from patient reviews. Experimental evaluation on the drug dataset, comprising patient reviews, demonstrates that the proposed hybrid sentiment classification model achieves 99.80% accuracy. It outperforms traditional machine learning and deep learning approaches. The proposed framework not only improves sentiment classification accuracy but also provides transparent and improved top-k drug recommendations, offering explainable patient-opinion-informed recommendation rankings derived from large-scale review data. The generated rankings are derived from patient-generated review data and should therefore be interpreted as opinion-aware recommendations rather than clinically validated treatment recommendations.

Managing Climate-Change Impacts through Predictive Analytics and AI-Driven Automated Irrigation

Gomathi S — 2026-09-01

Climate change poses a major challenge to the agricultural sector because it disrupts weather patterns and reduces water availability. Effective irrigation-water management and accurate climate prediction are essential for mitigating these effects and supporting sustainable farming. This study analyzes the combined use of convolutional neural networks (CNNs) and artificial intelligence (AI) to predict future climate variations and automate irrigation. The proposed system uses a CNN to analyze historical climate data, satellite imagery, and weather forecasts and to generate accurate regional climate predictions. These forecasts are then supplied to an AI-based irrigation system that allocates water according to predicted weather conditions, soil-moisture levels, and crop requirements. The AI system uses real-time sensor data and dynamically adjusts irrigation schedules to improve water-use efficiency and minimize waste. Experimental findings indicate that integrating CNN-based prediction with AI-driven control can improve water management under farming conditions by providing farmers with insight into climate variability and an automated mechanism for reducing water loss. The approach demonstrates a sustainable and scalable method for improving agricultural productivity while mitigating climate-change risks.

EPAD: Email Phishing Attack Detection Powered by Machine Learning and Natural Language Processing

Mansi Harihar — 2026-09-15

Phishing is a form of social engineering attack which takes advantage of human vulnerability to exploit users. Email phishing is an emerging and prevalent form of cyber attack that impersonates a trusted entity. In today's world, email has become a de facto mechanism of communication for individuals and businesses. Phishing attacks on emails are growing constantly. Hence, the need for email phishing attack detection is increasingly becoming crucial. This paper presents an email phishing attack detection (EPAD) system using Machine Learning and Natural Language Processing (NLP) techniques to detect phishing emails. The proposed work makes use of algorithms which include Support Vector Classifier, Naive Bayes, Logistic Regression, Random Forest and eXtreme Gradient Boosting (XGBoost) for the prediction of benign or phishing emails. The EPAD system focuses on content based and textual features in detecting phishing emails. Readability scores, sentiments and attachments related information are also considered for the effective phishing email detection. The EPAD system uses NLP techniques which include Term Frequency-Inverse Document Frequency (TF-IDF) and Word2Vec for capturing the semantics of the email messages. Explainable AI is employed to provide interpretability of results of email phishing detection. Consequently, the EPAD system offers the reasoning about the prediction of phished or benign email and thereby enhances user trust. Experimental outcomes demonstrate that XGBoost with Word2Vec performs optimally in comparison with other algorithms in respect of accuracy, F1-score, Area under Receiver Operating Characteristic Curve (ROC-AUC) and Matthews Correlation Coefficient (MCC) for the detection of phishing emails. A thorough evaluation carried out in terms of ablation study, statistical performance analysis and sensitivity analysis validates the effectiveness of the proposed system.

Early Depression Risk Screening using Leakage-Aware Ensemble Learning with Probability Calibration

Mohitsinh Parmar — 2026-09-22

Depression among university students is an important mental health concern, particularly in technical and engineering education environments characterized by academic workload, competition, and financial stress. Although machine learning approaches are increasingly used in student mental health research, many previous studies rely on leakage-prone or post-outcome variables, limited validation protocols, and discrimination-focused evaluation, which may reduce their suitability for realistic early screening settings. This study presents a leakage-aware and calibration-aware ensemble learning framework for early depression risk screening among technical education students by excluding leakage-prone predictors during model development. Random Forest, Gradient Boosting, and stacking-based ensemble models were evaluated using nested stratified cross-validation, pooled out-of-fold prediction analysis, isotonic probability calibration, threshold sensitivity analysis, subgroup robustness assessment, and SHAP-based interpretability analysis. Experiments conducted on a publicly available dataset filtered to technical degree programs (N = 7,807) showed stable cross-validated performance, with the calibrated stacking ensemble achieving ROC–AUC = 0.8718 ± 0.0081, PR–AUC = 0.8888 ± 0.0078, and recall = 0.8408 ± 0.0100. Statistical testing showed no significant performance difference between stacking and calibrated Gradient Boosting, suggesting that simpler calibrated models may provide comparable screening performance in this dataset. Leakage-ablation analysis showed that inclusion of the post-outcome suicidal-thoughts variable increased pooled out-of-fold ROC–AUC from 0.8701 to 0.9224, highlighting the importance of leakage-aware feature selection for realistic evaluation. A reduced-feature screening model using six early-available predictors also maintained competitive performance (ROC–AUC = 0.8616), supporting the feasibility of lightweight institutional screening. SHAP stability analysis demonstrated consistent feature rankings across validation folds, with academic pressure, financial stress, work/study hours, and sleep duration identified as influential predictors. Overall, the proposed framework provides an interpretable and methodologically transparent approach for depression risk prioritization in educational settings while reducing performance inflation associated with leakage-prone features.

Adaptive Komodo Mlipir–Optimized Spatiotemporal Graph Neural Network for Dataset-Specific Physiological-State Classification

Kishore Kanna R — 2026-08-10

Physiological-state classification requires models that can represent dependencies among sensor channels while preserving temporal variation. This study presents a dataset-specific evaluation of an Adaptive Komodo Mlipir Algorithm-optimized spatiotemporal graph neural network (AKMA-ST-GNN) using four public physiological datasets with different modalities and target definitions: STEW for cognitive workload, SEED for emotion, DEAP for affect, and WESAD for stress and amusement. The datasets were analyzed independently. No EEG signal from one dataset was combined with EMG or peripheral signals from another dataset, and workload, emotion, affect, and stress labels were not collapsed into a common target. The proposed workflow included signal filtering, training-set standardization, frequency-domain characterization where physiologically appropriate, within-dataset channel graph construction, spatiotemporal graph learning, and dataset-specific classification. AKMA was used as a hyperparameter-search wrapper rather than as an additional predictive layer. Under the preliminary 80% training and 20% testing protocol preserved in the source manuscript, the reported accuracies were 98.60% for STEW, 98.31% for WESAD, 98.06% for DEAP, and 98.12% for SEED. Precision, F1-score, specificity, and sensitivity were available only for selected datasets. Prediction-level outputs, class supports, confusion matrices, and decision scores were not retained, therefore, macro F1, weighted F1, balanced accuracy, Matthews correlation coefficient, Cohen's kappa, and ROC/AUC could not be reconstructed. Repeated-seed results and fold- or participant-level scores were also unavailable, preventing calculation of standard deviations, confidence intervals, and statistical significance tests. The reported values are consequently interpreted as preliminary evidence of technical feasibility rather than proof of stable or subject-independent generalization.

Evaluation of Structural, Optical, Morphological and Colloidal Stability of NiO Nanoparticles Synthesized Through Green Method

Leninbarathi K — 2026-09-09

NiO nanoparticles have been produced by an eco-friendly combustion method utilizing water-based extracts from several parts of the Eichhornia crassipes plant, specifically the root, stem and leaf. Nickel nitrate hexahydrate and urea (fuel) are taken as precursors while plant extracts operated as natural stabilizers and reducing agents. We used a lot of different analytical tools to make sure that the nanoparticles we made were fully characterized. These tools included zeta potential analysis, field emission scanning electron microscopy (FESEM), photoluminescence (PL), X-ray diffraction (XRD) and Fourier transform infrared spectroscopy (FTIR). The XRD analysis confirmed that the NiO nanoparticles have a single-phase cubic structure with higher crystalline nature. The average crystallite sizes of the samples are in the range of 13 to 17 nm. The vibrational analysis of NiO nanoparticles revealed the presence of Ni-O stretching vibrations between 400 to 600 cm^-1. The optical band gap values are found to be in between 3.40 and 3.37 eV. The photoluminescence spectra showed too much ultraviolet and violet light, which was explained by intrinsic transitions and defect states, including oxygen vacancies. FE-SEM micrographs revealed that most of the nanoparticles are spherical and coalesced together in different ways depending on the plant extract employed. Tests of zeta potential showed that the colloids were pretty stable. This was especially true for the NiO nanoparticles that came from the stem. The results reveal that extracts from Eichhornia crassipes can be utilized to generate NiO nanoparticles with optical and structural properties that can be changed. The present analysis showed that the NiO nanoparticles are useful for photocatalysis, sensors, optoelectronics and energy devices.

A Hybrid CNN–LSTM Deep Learning Framework for Automated Malware Classification using Spatial–Sequential Feature Fusion

Chevella Anil Kumar — 2026-09-22

The high rate of growth of advanced and highly obfuscated malware has made the use of conventional signature detection mechanisms less viable. In order to overcome this problem, this paper suggests a hybrid framework for malware classification based on the combination of Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks to learn spatial and temporal features simultaneously. The given model converts malware feature vectors into structured grayscale representations from which CNN layers extract spatial features, and sequential opcode-related patterns are learned by an LSTM branch. Aspects of both domains are combined into a single embedding and used to classify nine malware families. Tests that have been carried out with a benchmark dataset show that the hybrid CNN-LSTM model attains a classification accuracy of 97.27% which is very high compared to baseline LSTM-only and CNN-only frameworks. The model demonstrates strong generalization, with weighted precision, recall, and F1-score values above 97% and highly discriminative ROC-AUC scores (reaching 1.000 for major classes). An in-depth analysis, such as confusion matrix analysis, precision-recall curves, and a comparison with the state-of-the-art methodologies demonstrate that the proposed architecture achieves performance comparable to that of leading malware detection models reported in recent literature. These findings validate the power of hybrid feature fusion to capture both static and dynamic behavioral traits of malware to provide a powerful, scalable, and highly accurate solution for next-generation cybersecurity systems.