Science, Technology, and Communication Journal

A hybrid multi-corpus RAG framework for faithful and source-grounded MUI fatwa question answering

2026-08-13T10:55:16+07:00

The increasing use of large language models (LLMs) for Islamic legal information retrieval presents substantial challenges related to hallucinations, source relevance, and the limited coverage of single-corpus retrieval-augmented generation (RAG) systems. This study proposes a sequential multi-corpus RAG architecture for a hybrid MUI fatwa chatbot that prioritises authoritative fatwa documents while providing controlled fallback retrieval from a curated hadith corpus. The system employs dense semantic retrieval using FAISS for the primary MUI fatwa corpus, followed by similarity threshold filtering and LLM-based relevance verification. When no sufficiently relevant fatwa context is identified, the query is redirected to sparse retrieval of over 12,370 hadith records from Sahih al-Bukhari and Sahih Muslim. The proposed system was compared with a conventional single-corpus RAG configuration using the BERTScore and Giskard’s RAG evaluation toolkit (RAGET). BERTScore evaluation on ten questions showed that the proposed architecture increased the average F1-score from 0.6664 to 0.7546, representing a relative improvement of 13.23%. On 41 successfully paired evaluation records, the Giskard faithfulness score increased from 0.8537 to 0.9439, equivalent to a 10.57% relative improvement. However, both systems produced an identical abstain rate of 26.83%, indicating that the proposed architecture primarily improved answer relevance and contextual grounding rather than overall question coverage. These findings demonstrate that hierarchically organising knowledge sources according to their authority and function can enhance the reliability of RAG-based Islamic legal information systems. Nevertheless, the chatbot should be regarded as a source-grounded information-access tool and not as a substitute for fatwas or qualified Islamic scholars.

Explainable AI (XAI) for feature selection optimization in deep neural networks using the Howells craniometric dataset

2026-08-11T11:04:10+07:00

Biological profile identification through cranial metrics is a crucial procedure in forensic anthropology. The use of high-dimensional tabular craniometric data often triggers information redundancy problems that degrade the performance of computational models. Moreover, rejection in the medical and legal domains frequently occurs due to the black-box nature of Deep Learning algorithms, which fail to provide anatomical transparency. This research develops an advanced hybrid expert system framework for classification using craniometric data. In the pre-processing stage, feature selection is optimized using a hybrid approach of genetic algorithm (GA) and particle swarm optimization (PSO) to optimally extract 15 critical cranial parameters from the 82 raw features of the William W. Howells craniometric dataset. In the classification stage, a 1D convolutional neural network (1D-CNN) is constructed. To validate the effectiveness of the proposed method, the GA-PSO 1D-CNN model is compared against a baseline 1D-CNN model trained on all 82 raw features without feature selection. To overcome the transparency issue, the black-box nature of the model is dismantled using the explainable AI (XAI) method based on SHAP via GradientExplainer. The experimental results show that the GA-PSO optimization successfully reduces feature dimensionality deterministically, preventing the curse of dimensionality while maintaining classification precision. Furthermore, the visual SHAP analysis is able to present the hierarchy of the 15 most determining craniometric parameters, providing a rationalization of operational decisions that are cross-validated with the principles of human biological anatomy.

High-energy supercapacitor based on phosphoric acid (H3PO4)-activated Moringa oleifera leave waste as the electrode material

2026-10-22T11:06:14+07:00

Moringa oleifera-leaves were used as the main starting material for making activated carbon for supercapacitor electrodes. The research focused on producing Moringa oleifera-leaves activated carbon and examining its physical properties and electrochemical performance. The activated carbon was produced through a chemical activation process using 0.5 M H₃PO₄, carbonization under N₂ gas, and physical activation using CO₂ gas. Density measurements showed a decrease from 0.94 g/cm³ to 0.60 g/cm³ following pyrolysis, representing a reduction of 36.17%. Electrochemical testing was performed using cyclic voltammetry (CV) and galvanostatic charge–discharge (GCD) methods in a 1 M Na₂SO₄ electrolyte with a current density of 1 A/g. Results obtained within a test voltage window of 0 – 1 V yielded specific capacitances of 45 F/g (CV) and 130 F/g (GCD). Further testing across a wider operating voltage window of 0 – 2 V resulted in a specific capacitance of 153.33 F/g, with a specific energy of 85.18 Wh/kg, and specific power of 1,840 W/kg. These findings demonstrate the potential of Moringa oleifera leaves waste for use as an electrode material in supercapacitors capable of operating at a voltage range of 0 – 2 V, thereby supporting the development of high-energy supercapacitors.

Formulation and effectiveness testing of mouthwash (Gargarisma) combination of guava leaves (Psidium guajava L. folium) and bay leaves (Syzygium polyanthum folium) extracts as an antibacterial agent against Streptococcus mutans

2026-10-11T11:50:22+07:00

The dental caries disease is the result of the bacterial infection by Streptococcus mutans bacteria which cause the decay of the teeth. The extraction of guava leaves (Psidium guajava L. folium) and bay leaves (Syzygium polyanthum (Wight) Walp.) has secondary metabolites with antibacterial action. This research is intended to investigate the effect of combination of both plant extracts as the inhibitors of Streptococcus mutans by making the mouthwash in the three concentrations of F1, F2, and F3. Plant identification, phytochemical testing, physical testing, and disc diffusion test are done in order to identify the identity of the plants, metabolite content of plant, physical quality of the formulation, and antibacterial activity of the formulation. Statistical analysis used for the inhibition zone diameter (IZD) value is one-way ANOVA. From the phytochemical test result, both of plant extracts have secondary metabolites of alkaloid, flavonoid, tannin, and phenol. Physical quality of the F1, F2, and F3 formulations remained stable until the four weeks of storage period. Antibacterial assays revealed that F3 exhibited the highest inhibitory activity against Streptococcus mutans with an IZD of 8.67 ± 0.764 mm, followed by F2 at 7.67 ± 0.577 mm and F1 at 4.33 ± 3.786 mm. One-way ANOVA demonstrated significant differences in IZD between groups (F = 17.135; p < 0.001), following the order of F3 > F2 > F1. Post hoc analysis using the Games-Howell test showed that F2 and F3 had significantly higher antibacterial activity than the negative control. However, the differences in effectiveness among F1, F2, and F3 were not statistically significant.

The impact of ADASYN on Alzheimer's disease classification performance using backpropagation neural networks

2026-10-24T13:27:20+07:00

Alzheimer's disease (AD) is a neurodegenerative disorder that causes a decline in cognitive ability and the capacity to perform daily activities. Class imbalance in Alzheimer's data may bias machine learning models toward the majority class and reduce their ability to recognize the minority class. This study applied adaptive synthetic sampling (ADASYN) to a backpropagation neural network (BPNN) to balance the class distribution and improve the classification of Alzheimer’s disease. The Kaggle dataset contained 2,149 records (760 Alzheimer and 1,389 non-Alzheimer) with 34 input variables, which were reduced to 32 after selection, and one binary output variable. After preprocessing, the data were split using 10-fold cross-validation. Z-score normalization and ADASYN were applied to the training data of each fold, whereas the test data received only Z-score normalization. The 32-33-1, 32-48-1, and 32-63-1 architectures were evaluated with learning rates of 0.01, 0.05, and 0.09, each with and without ADASYN. The 10-fold average results showed that ADASYN consistently improve neither accuracy (80.47% with ADASYN versus 80.41% without) nor recall, which was higher in only 2 of 9 configurations (70.83% versus 72.28%). The best configuration was the 32-63-1 architecture with a learning rate of 0.01 and ADASYN, which achieved 81.57% accuracy, 75.05% precision, 71.84% recall, and 73.31% F1-score. ADASYN tended to reduce false positives but did not reduce undetected Alzheimer cases (false negatives) providing no better results for Alzheimer’s disease diagnosis.

Unsupervised federated autoencoders for IoT botnet detection under non-IID and contaminated training data

2026-10-13T10:21:17+07:00

Federated learning (FL) has been proposed for privacy-preserving intrusion detection in internet of things (IoT) networks, yet most studies assumed labeled traffic and identically distributed client data. This study benchmarked unsupervised federated autoencoders for IoT botnet detection under non-IID heterogeneity and training-data contamination. N-BaIoT and Edge-IIoTset were partitioned into clients using IID, natural per-device, and Dirichlet partitions with concentration parameters of 1.0, 0.5, and 0.1, and botnet traffic was injected into the benign training data of 30% of the clients at rates of 0% – 20%. Lightweight autoencoders were trained on unlabeled traffic and aggregated with FedAvg and FedProx, with local-only and centralized training as references, in 264 runs over five random seeds. FedAvg did not outperform local-only training in any of the ten dataset–partition settings (mean AUC-PR difference of 0.058, Holm-adjusted p = 0.014) and, on N-BaIoT, declined by up to 0.087 after an early peak. FedProx prevented this decline on N-BaIoT (AUC-PR up to 0.999) but gave no gain on Edge-IIoTset. Under heterogeneity, a single global threshold raised the false positive rate from about 1% to 22.2%, whereas client-level thresholds kept it at or below 1.0%. Contamination of 20% reduced the worst-client F1-score by 0.33 for FedAvg and 0.47 for FedProx, and client-level thresholds did not recover this loss. These findings indicated that standard aggregation and a single global threshold were insufficient for unsupervised federated botnet detection and motivated label-free, heterogeneity-aware aggregation with client-level threshold calibration.

Comparative analysis of feature selection methods for support vector machine-based sex classification using craniometric data

2026-09-30T09:37:05+07:00

Sex classification is a fundamental stage in forensic anthropology that can be supported by machine learning to achieve objective biological profiling. To address the high dimensionality and inter-trait multicollinearity of craniometric measurements, this study introduces a comparative feature-optimization framework that evaluates two distinct filtering paradigms entropy-based information gain (IG) and redundancy-aware correlation-based feature selection (CFS) integrated with a support vector machine (SVM) classifier against a full-feature baseline model. Experiments were conducted using the William W. Howells craniometric dataset, comprising 2,524 human crania (1,368 males and 1,156 females) characterized by 82 osteometric predictor variables after removing administrative identifiers. Model performance on a 70:30 training-testing split was evaluated using accuracy, precision, recall, F1-score, and the Matthews correlation coefficient (MCC). Experimental results show that the baseline SVM using all 82 features achieved an accuracy of 81.10% and an F1-score of 0.8110. Integrating feature selection substantially improved classification performance while compressing data dimensionality. The IG–SVM model achieved the highest accuracy of 86.73% (F1-score = 0.8674, MCC = 0.7348) using 20 selected features (a 75.60% feature reduction). Meanwhile, the CFS–SVM model achieved a virtually identical accuracy of 86.69% (F1-score = 0.8670, MCC = 0.7335) while requiring only 10 features (an 87.80% feature reduction). Crucially, these findings demonstrate that eliminating 75.60% to 87.80% of the original craniometric features not only enhances SVM classification accuracy by more than 5.5 percentage points by removing noisy and collinear variables, but also substantially reduces osteometric measurement complexity. Consequently, the parsimonious 10-feature CFS–SVM configuration provides an efficient and interpretable framework for computer-assisted sex estimation in forensic anthropology.

AI-based feature selection for craniometric sex classification: A nested cross-validation study on the Howells dataset

2026-10-05T13:00:15+07:00

Sex estimation is an important part of forensic biological-profile reconstruction. However, machine-learning results can be overly optimistic when feature selection and hyperparameter tuning are performed without separating them from model evaluation. This study compared five feature-selection methods, namely Information Gain, ReliefF, LASSO, SHAP-based ranking, and support vector machine recursive feature elimination (SVM-RFE), using five classifiers: support vector machine (SVM), random forest, logistic regression, decision tree, and multilayer perceptron. The study used 2,524 individuals from a Howells-derived craniometric dataset, consisting of 1,156 females and 1,368 males. A stratified hold-out set containing 20% of the data was reserved before model development. After missingness filtering based on the development data, 71 predictors remained. Model development used 5-fold outer and 5-fold inner cross-validation, with 3, 5, 7, 10, and 15 candidate features. SVM-RFE combined with SVM obtained a nested-cross-validation accuracy of 0.8767 ± 0.0098, with an F1-score of 0.8852 and ROC-AUC of 0.9481. After retuning using the full development set, the final 15-feature model achieved 0.8970 accuracy, 0.9055 F1-score, 0.9511 ROC-AUC, 0.9088 sensitivity, and 0.8831 specificity on 505 internal hold-out observations. The AI-oriented selectors had a mean accuracy only 0.0074 higher than the filter selectors, which was below the predefined 0.05 improvement threshold. Classifier choice had a significant effect on accuracy, while the feature-selection method and its interaction with the classifier were not significant. SHAP analysis identified ZYB, AUB, and MDH as the three highest-ranked features. The results show that a relatively small number of craniometric features can provide good classification performance under the evaluation procedure used in this study. However, the differences between feature-selection approaches were relatively small, and further validation on independent populations and data collected at different time points is needed before the model can be considered for forensic use.

Islamic law-based knowledge management framework for divorce mediation in religious courts

2026-10-19T08:50:41+07:00

This study develops a conceptual knowledge management framework informed by Islamic law to support divorce mediation in Indonesian Religious Courts. It employs qualitative secondary analysis of dissertation findings from five Class IB Religious Courts within the jurisdiction of the Padang Religious High Court, aggregate mediation data for 2020–2024, and literature on knowledge management, organizational learning, digital transformation, and AI governance. Of 18,033 registered divorce cases, 3,065 (17.00%) were recorded as undergoing mediation. Among these mediated cases, 1,504 (49.07%) were unsuccessful, 485 (15.82%) were fully successful, 894 (29.17%) were partially successful, 68 (2.22%) were not conducted, and 114 (3.72%) remained in progress. The findings highlight the need to strengthen knowledge documentation, mediator learning, institutional knowledge sharing, and responsible digital knowledge use. In response, the article proposes six components of the Islamic law-informed knowledge management framework for religious Courts (IKM-RC): knowledge identification and creation; knowledge capture and codification; validation and digital repository; knowledge sharing and organizational learning; AI-assisted knowledge utilization; and evaluation and continuous improvement. The framework integrates sulh and maqasid al-shari'ah as normative orientations, alongside confidentiality, source traceability, accountability, and human oversight. IKM-RC is a conceptual proposal rather than a validated intervention; expert review, prototype development, and empirical evaluation are required before its practical effectiveness can be assessed.

Investigations of normalization strategies on [18F]-FDG pet radiomics feature: A multi-cancer analysis

2026-10-21T09:45:25+07:00

This study investigated the effects of normalization methods on the robustness of [¹⁸F]-FDG PET radiomic features across different fixed bin number (FBN) settings and cancer types. A total of 107 radiomic features were evaluated in melanoma, non-small cell lung cancer (NSCLC), and lymphoma datasets across 32 FBN settings ranging from 8 to 256 in increments of 8. Features were assessed without normalization and after Z-score, Min–Max, or Robust scaling, applied independently at each FBN setting. The intraclass correlation coefficient (ICC) was used as the measure of agreement across discretization settings, while the coefficient of variation (CV) was examined as a measure of variability. Normalization effects varied across radiomic feature classes and cancer datasets. Z-score and Robust scaling generally increased ICC for many texture features compared with no normalization, whereas the response to Min–Max scaling was more variable. CV distributions also differed across normalization methods and cancer types, with Z-score producing the highest mean CV and Min–Max the lowest in all three datasets. However, between-method differences in CV were interpreted descriptively, as normalization altered the feature means underlying its calculation and therefore limited direct comparability. Post extraction normalization was associated with increased ICC-based agreement of [¹⁸F]-FDG PET radiomic features across FBN settings, with Z-score normalization showing the most consistent results across cancer cohorts. However, the effects varied by feature and normalization method, highlighting the need for feature-level assessment and transformation-aware interpretation of variability.

Characterizing alpha-band electroencephalography activity from cognitive listening test in undergraduate students

2026-08-27T11:37:24+07:00

This study examined alpha-band electroencephalography (EEG) activity during a cognitive listening test in undergraduate students aged 19 – 22 years. EEG signals were recorded using an EMOTIV EPOC+ device during resting-state conditions and a listening-based cognitive test involving synonyms, antonyms, word analogies, premise-based reasoning, and narrative-audio comprehension tasks. Alpha-band power spectral density (PSD) was estimated using Welch's periodogram method to describe the magnitude and scalp distribution of alpha-band activity in seven undergraduate students (aged 19 – 22 years) during resting state and cognitive task condition. The resting-state results showed consistently higher alpha PSD during eyes-closed than eyes-open conditions, both before and after the cognitive listening test. During task performance, scalp topographies showed substantial inter-individual variation in the magnitude and spatial distribution of alpha band PSD. This descriptive pattern cannot be attributed to specific cognitive processes or strategies because behavioral measures and inferential statistics were not available. The findings indicate that Welch based PSD analysis can be used to describe EEG activity during listening tasks, larger samples, behavioral measures, artifact correction, and statistical comparisons are required before any EEG markers of listening performance can be proposed.

Adaptive and explainable machine learning for Indonesian SMS phishing detection: Few-shot prototype learning, simulated concept drift, and SHAP/LIME feature analysis

2026-10-05T13:06:57+07:00

Phishing delivered through short text messages evolves quickly, producing scarce and imbalanced labelled data and shifting lexical patterns that challenge conventional supervised detectors. This study develops and audits a transparent detection framework on a private corpus of 15,950 Indonesian-language SMS messages, of which 13,411 unique messages (8,091 phishing, 5,320 normal) remain after removing 3,771 normalized duplicates. A class-weighted Logistic Regression over TF-IDF unigram–bigram features serves as the primary detector, and it is subjected to a battery of scenario evaluations: stratified five-fold cross-validation, a 2-way one/five/ten-shot episodic protocol, a controlled simulated lexical-drift stream with adaptive and static online learners, a binary class-introduction test, and an explainability pipeline built from exact linear SHAP and perturbation-based LIME. The detector attained an accuracy of 0.9806, an F1-score of 0.9838, a Matthews correlation coefficient of 0.9598, a ROC-AUC of 0.9973, and a PR-AUC of 0.9983 on a stratified hold-out. Five-fold cross-validation confirmed consistent performance with F1 scores ranging from 0.9829 to 0.9888. Episodic evaluation showed monotonic improvement from the one-shot to the ten-shot setting. Under simulated lexical drift, an adaptive online learner preserved a post-drift F1 of 0.9808 against 0.9604 for a static learner. Exact linear SHAP reproduced the decision function to within 1.8 × 10^-15, LIME achieved a local fidelity of R² = 0.921, and SHAP-based selection retaining 500 of 10,000 features (a 95% reduction) reduced F1 only modestly to 0.9741 with a mean pairwise Jaccard stability of 0.818. The findings indicate that a compact, class-weighted linear model can deliver strong, stable, and auditable SMS phishing detection, while few-shot and drift experiments delineate the boundaries within which limited-support and non-stationary performance should be interpreted.

Electrocardiogram signal analysis using fast Fourier transform method for heart rate diagnosis

2025-10-24T14:07:22+07:00

This study aims to analyze electrocardiogram (ECG) signals using the fast Fourier transform (FFT) method to support heart rate diagnosis. The data used consists of nine recordings of patient ECG signals obtained in graphic form, then converted into numerical data using PlotDigitizer software. Furthermore, the data was analyzed in the frequency domain using the FFT algorithm formulated through MATLAB software. The transformation results showed that the dominant frequency of each ECG signal ranged from 1.0332 Hz to 1.6172 Hz. A comparison between the FFT results and normal ECG signals was conducted to evaluate the suitability of the waveform and rhythm. Quantitative evaluation was conducted using the mean squared error (MSE) method to measure the level of difference between signals. The results confirm that the frequency-based method significantly contributes to more accurate measurements and greater efficiency in heart rate diagnostics. This method also supports the advancement of sophisticated techniques in biomedical signal processing.

Synthesis of hybrid nanoparticle for photocatahlyst, antibacterial, and anticancer applications: A review

2025-10-24T14:10:05+07:00

Hybrid nanoparticles have attracted considerable interest in recent years owing to their multifunctional potential across a spectrum of applications, encompassing photocatalysis, antibacterial agents, and anticancer therapeutics. This review furnishes a comprehensive survey of hybrid nanoparticle synthesis, with a focus on diverse synthetic methodologies, constituent materials, and pertinent applications. A range of synthesis techniques, including co-precipitation, hydrothermal methods, mechanical milling, electron beam lithography, gas phase deposition, electrical explosion of wires, spray pyrolysis, laser pyrolysis, thermal decomposition, solvothermal synthesis, sol-gel processes, polyol methods, non-plasma thermal techniques, microemulsion synthesis, vapor deposition and patterning, as well as biological (green) synthesis routes, are explored, with a critical appraisal of their respective merits and limitations. Commonly employed nanoparticulate materials, including metal oxides (e.g., TiO₂, ZnO, AgO, CuO, Fe₂O₃), metallic nanoparticles (e.g., Au, Ag), and organic materials (e.g., plant extracts), are discussed in detail. This review also addresses the applications of hybrid nanoparticles in photocatalysis for the degradation of wastewater pollutants, as antibacterial agents for the control of microorganisms, and in anticancer therapies. Furthermore, the underlying mechanisms governing the photocatalytic, antibacterial, and anticancer activities of hybrid nanoparticles are also elucidated. By synthesizing recent advancements in the synthesis and applications of hybrid nanoparticles, this review aims to provide valuable insights and stimulate further investigation within this rapidly evolving domain.

Multi-level cascade fiber Bragg grating spectral simulation with apodization variations

2026-01-14T15:38:40+07:00

This paper presents a numerical investigation of the spectral characteristics of multi-level cascade fiber Bragg grating (FBG) structures employing Gaussian, uniform, and tangent hyperbolic apodization profiles. The analysis focuses on the combined influence of apodization function and cascade levels (4, 6, and 8 gratings) on Bragg wavelength shift, reflectivity stability, optical power distribution, and strain sensitivity. Simulation results demonstrate that Gaussian apodization provides superior spectral smoothness and wavelength stability across all cascade configurations, while tangent hyperbolic apodization exhibits the highest sensitivity to mechanical perturbation and maintains robust spectral integrity at higher cascade levels. In contrast, uniform apodization suffers from pronounced sidelobe effects, leading to spectral degradation as the number of cascaded gratings increases. These findings highlight the critical role of apodization selection in optimizing multi-level cascade FBG sensors for stable and sensitive multipoint sensing applications.

Analysis of the effect of mechanical bending on the optical power of Bragg grating optical fibers in thermal measurements

2026-01-14T15:40:13+07:00

Bending of optical fibers is employed as a control parameter to evaluate the thermo-mechanical sensitivity and response of fiber Bragg grating (FBG) sensors in temperature measurements on curved structures. A uniform FBG is operated at two wavelengths, 1310 nm and 1550 nm, with bending angles from 0° to 60° and exposure to hot water vapor within a temperature range of 35°C – 49°C. The output power is measured using an optical power meter (OPM), and the Bragg wavelength shift is obtained from the transmission response. At 1550 nm, the transmitted power decreases from -25.83 dBm (0°) to -25.97 dBm (60°), accompanied by increased Bragg wavelength shifts at larger angles and higher temperatures. In contrast, at 1310 nm, the power variation is smaller (-30.57 dBm to -30.71 dBm), and Bragg wavelength remains almost unchanged at lower temperatures, with oscillations becoming noticeable only at larger bending angles. The 0° – 20° range exhibits the most stable response, whereas the 40°– 60° range at 1550 nm demonstrates the highest sensitivity to temperature variations. Overall, operation at 1550 nm is more responsive to thermomechanical effects, while 1310 nm exhibits greater stability against fiber bending.