2692-4048ISSN (Online)
2770-0070ISSN (Print)
Finding and treating different types of strabismus, which is when the eyes do not line up properly, can be challenging. This study introduces a deep learning system that automatically identifies five types of strabismus: esotropia, exotropia, hypertropia, hypotropia, and normal eye alignment. It combines EfficientNet-B0 with Grad-CAM to improve how the system recognizes and classifies these conditions accurately. These help EfficientNet-B0 improve how it picks out important features using squeeze-and-excitation blocks, which capture key details needed for accurate classification. Grad-CAM further refines this process and localizes the critical regions in the feature maps more effectively to improve interpretability. We trained the model on a dataset of 10,000 balanced images across the five classes, achieving a classification accuracy of 99.43% and 96.33% for training and testing data, respectively. The model's focus-based architecture ensures that clinicians' set goals are met in terms of the model's efficiency and reliability for predictions.
Read MoreDoi: https://doi.org/10.54216/FPA.210101
Vol. 21 Issue. 1 PP. 01-12, (2026)
Particulate Matter (PM) concentration significantly affects public health, exacerbating respiratory conditions and contributing to environmental challenges. This study presents a real-time Internet of Things (IoT)-based portable particulate matter monitoring device utilizing the PMS5003 sensor. The device measures PM1.0, PM2.5, and PM10 concentrations and uploads the data to the cloud at 15-second intervals for real-time visualization. A two-week observational study in South Tangerang, Indonesia, revealed peak PM2.5 and PM10 levels of 218 µg/m³ and 232 µg/m³, respectively, on weekdays, compared to a weekend low of 19.76 µg/m³ for PM2.5. Variations were influenced by anthropogenic factors, including vehicular and industrial activity. Data analysis showed a 78% reduction in PM2.5 levels during weekends, highlighting the impact of human activity on air quality. These findings underscore the impact of anthropogenic activities on air quality and demonstrate the effectiveness of IoT-based systems in environmental monitoring. The study highlights the potential for such technology to support data-driven strategies for pollution management and public health improvement.
Read MoreDoi: https://doi.org/10.54216/FPA.210102
Vol. 21 Issue. 1 PP. 13-26, (2026)
To increase scalability, response speed, and fault tolerance, modern web systems must have load balancing and caching solutions. Better resource allocation and traffic management control help to prevent system overload. This is essential to satisfy the growing need for perfect digital experiences. This work intends to demonstrate an adaptive load balancing system using real-time job scheduling, predictive analytics, and multi-layer caching, integrating artificial intelligence technology. Our hybrid deep learning and storage systems lower data retrieval time and estimate traffic. This approach tremendously increases the efficiency of online systems. Unlike conventional load balancing systems, which rely on either static or rule-based traffic distribution, our approach employs artificial intelligence-based dynamic allocation to real-time resource adjustment. Our solution forecasts workload surges and pre-allocated resources suitably using deep neural networks in conjunction with past traffic data. To hasten data retrieval, the multi-layer caching approach makes use of content delivery networks (CDNs) and cloud-based storage. This lessens the double effort required and helps one discover objects more easily. Among the several advantages, the new approach offers over the old ones are a 40% decrease in energy use, a 20% improvement in resource use, and a 50% improvement in reaction time. This approach has exceeded round robin and dynamic load balancing in actual AWS simulations. These findings highlight how incorporating predictive analytics driven by artificial intelligence might improve current site designs. For cloud platforms, IoT systems, and high-traffic online applications needing efficiency and fast adaption, this approach performs well.
Read MoreDoi: https://doi.org/10.54216/FPA.210104
Vol. 21 Issue. 1 PP. 45-62, (2026)
This study aims to forecast the export volumes of oil and gas and non-oil and gas sectors in Indonesia, as export volumes reflect the economic condition of a country. The research utilizes data from BPS, spanning from January 2018 to December 2023, and employs the Fuzzy Time Series (FTS) methodology. Six different methods are applied: First-Order FTS Chen, First-Order FTS Cheng, Second-Order FTS Chen, Second-Order FTS Cheng, Markov Chain FTS, and Time-Invariant FTS. FTS is a predictive technique based on fundamental logic and various concepts and rules within fuzzy sets. The prediction accuracy is evaluated using the Mean Absolute Percentage Error (MAPE). The MAPE values for these six methods are compared to determine the most suitable method for this case study. The findings reveal that First-Order FTS Chen achieves an accuracy of 4.07%, First-Order FTS Cheng 4%, Second-Order FTS Chen 1.61%, Second-Order FTS Cheng 1.58%, Markov Chain 3.96%, and Time-Invariant 8.88%. The results indicate that Second-Order FTS Cheng provides the highest accuracy and is effective for predicting the export volumes of oil and gas and non-oil and gas sectors in Indonesia.
Read MoreDoi: https://doi.org/10.54216/FPA.210103
Vol. 21 Issue. 1 PP. 27-44, (2026)
The integration of blockchain technology and metaheuristic optimization has transformed healthcare systems by improving security, scalability, and data interoperability. Blockchain ensures decentralization, immutability, and privacy, making it a viable solution for electronic medical records (EMRs) and secure healthcare data management. Meanwhile, metaheuristic algorithms optimize blockchain networks by enhancing transaction efficiency, consensus mechanisms, and real-time medical data processing. This paper systematically reviews recent advancements in blockchain and metaheuristics for healthcare applications. We discuss existing privacy-preserving models, AI-driven optimization techniques, and hybrid consensus mechanisms, addressing their strengths and limitations. Through a structured methodology, we analyze research trends, security challenges, and computational bottlenecks. This study encompassed 300 research articles from nine global databases. Then, inclusion and exclusion criteria were applied, leading to the exclusion of 144 studies and the retention of 156 studies. Subsequently, quality assessments were conducted, resulting in the final inclusion of only 8 studies for data extraction. A three-phase methodology was followed: planning, conducting, and reporting. The studies covered the period from January 2020 to January 2025, and 10 evaluation questions were used to assess the quality of the studies. Our findings reveal that while blockchain enhances data security and interoperability, metaheuristic-driven AI further optimizes system efficiency. However, challenges such as scalability constraints, energy consumption, regulatory compliance, and AI-based cyber threats remain significant. Future research should focus on developing lightweight blockchain architectures, quantum- resistant cryptographic models, and federated AI-enhanced security frameworks to address these issues. By leveraging advanced blockchain and AI-driven metaheuristics, healthcare systems can achieve greater resilience, efficiency, and adaptive security.
Read MoreDoi: https://doi.org/10.54216/FPA.210105
Vol. 21 Issue. 1 PP. 63-78, (2026)
The significance of the Intrusion Detection System (IDS) is due to its capability in detecting attacks over the network. The current paper proposes a new feature selection method for misuse intrusion detection systems based on RNA encoding, where the proposed method includes five steps. Firstly, the KDD-Cup99 dataset is used and then select random records are used for both training and testing. Secondly, RNA encoding to encode each possible value in the dataset into RNA characters. Thirdly, the keys and their locations are extracted by dividing the achieved RNA sequences from previous steps into blocks with different sizes, then finding the most repeated blocks, choosing them as keys, and storing their location. The next step is the proposed feature selection method based on the extracted keys and their locations, depending on the place of the key within the feature number. Finally, the Raita algorithm for matching to search for keys before and after the applied features selection method. In terms of IDS performance evaluation, experimental outcomes of the proposed feature selection method show the capability of optimizing the time complexity and metrics.
Read MoreDoi: https://doi.org/10.54216/FPA.210106
Vol. 21 Issue. 1 PP. 79-88, (2026)
Rice plant disease detection is crucial in agriculture to prevent crop loss and enhance productivity. Traditional manual inspection methods often lead to inaccuracies, delays in diagnosis, and excessive pesticide use. To address these challenges, this study proposes an Artificial Layered Fuzzy Neural Network-based African Vulture Optimization (ALFNN-AVO) algorithm for early and accurate detection of rice plant diseases. The proposed framework integrates multiple advanced techniques, including Cross Fusion former (CF former) for feature extraction, Squeeze Excitation (SE) fusion for enhancing feature representation, and Spatial Fuzzy C-Means (SPFCM) for precise segmentation of affected plant regions. Furthermore, an Artificial Layered Depth Separable Neural Network (ALDSNN) is employed for multi-class classification of rice plant diseases. The Differential Bitwise African Vultures Optimization Algorithm (DBAVOA) is introduced to optimize the hyperparameters, ensuring improved convergence and classification performance. Experimental results validate the efficiency of the proposed model, achieving an accuracy of 98.87% and an execution time of 0.09 minutes, outperforming existing methodologies. The findings demonstrate that the proposed framework offers a reliable and computationally efficient solution for real-time rice plant disease detection, contributing to sustainable agricultural practices.
Read MoreDoi: https://doi.org/10.54216/FPA.210107
Vol. 21 Issue. 1 PP. 89-109, (2026)
Emotion recognition using electroencephalogram (EEG) signals has become a pivotal area in affective computing, particularly within the context of human–computer interaction and game-based environments. This study aims to enhance the accuracy and robustness of EEG-based emotion classification by introducing a hybrid framework that combines the k-Nearest Neighbors (KNN) classifier with advanced metaheuristic feature selection techniques. Using the publicly available GAMEEMO dataset, which includes EEG recordings from 28 subjects engaged in four emotionally distinct computer games (boring, calm, horror, and funny), EEG data were acquired through a 14-channel Emotiv Epoc+ device and labeled using the Self-Assessment Manikin (SAM) scale. Baseline machine learning models including Support Vector Machine (SVM), Decision Tree (DT), Multi-Layer Perceptron (MLP), and KNN were evaluated, with KNN achieving the highest base line performance. The KNN classifier was further optimized using several metaheuristic algorithms—namely WAO, BBO, GWO, GA, FA, PSO—and the proposed Improved Human Optimization Algorithm (iHOW). Experimental results show that the iHOW+KNN model achieved the best overall performance with an accuracy of 96.85%, sensitivity of 95.50%, specificity of 95.82%, and F1-score of 95.54%. Visual assessments using heatmaps, radar plots, and confidence intervals further validated the model’s reliability. These findings demonstrate the effectiveness of the iHOW+KNN framework in addressing the challenges of high-dimensional EEG data and highlight the potential of wearable EEG devices for real-time emotion recognition in affective computing applications into user experiences within the gaming environment.
Read MoreDoi: https://doi.org/10.54216/FPA.210108
Vol. 21 Issue. 1 PP. 110-130, (2026)
Finding jobs in today’s world is similar to finding a needle in the haystack. The modern job-search platforms present a language barrier for native-speakers and inexperienced candidates, making it difficult for them to compete in the job search race. NaukariCraft, a bilingual (Hindi & English) job search platform makes it easy for users to look for jobs, gain industrial insights, save time by finding relevant jobs tailored to skills and resume, building ATS friendly resume, and ATS score analyzer. NaukariCraft provides full guidance to novel applicants helping them find direction towards jobs tailored to their resume. Using advanced technology like Large Language Model (LLMs) and agents, NaukariCraft enhances user experience, improves employability through resume analysis, and reduces application fatigue. This paper outlines the methodology, proposed work, result, conclusion and future development avenues for NaukariCraft.
Read MoreDoi: https://doi.org/10.54216/FPA.210111
Vol. 21 Issue. 1 PP. 155-164, (2026)
As digital devices continue to process ever-increasing volumes of complex data, ensuring accurate and efficient machine learning performance has become a significant challenge. Traditional ensemble learning methods often attempt to address these issues through data sampling or partitioning; however, such approaches can introduce biases and fail to fully capture the underlying structure of the data. To address these limitations, this paper proposes a novel classification framework that integrates clustering with adaptive weighting strategies. The process begins by dividing the training data into clusters, each representing a specific subset of the overall data distribution. Separate machine learning models are then trained on these clusters, allowing each model to specialize in different areas of the data. When analyzing a test instance, its relationship to the individual clusters is evaluated using two key measures: the correlation coefficient, which assesses feature similarity, and the Mahalanobis distance, which calculates the statistical proximity to the cluster center. These values are subsequently used to generate optimized weights that determine the influence each model should have in the final ensemble prediction. By aligning model contributions with the structural similarities between the test and training data, the proposed approach enhances both the reliability and precision of classification. Experimental results demonstrate that this cluster-aware ensemble consistently outperforms both baseline and advanced classifiers on benchmark datasets.
Read MoreDoi: https://doi.org/10.54216/FPA.210109
Vol. 21 Issue. 1 PP. 131-141, (2026)
Remote sensing image evaluation faces continual challenges in extracting discriminative capabilities from complex; multi-scale landscapes the use of conventional spectral-spatial techniques, which often fail to capture hierarchical structures correctly. This examine proposes a brand-new methodology that leverages the discrete wavelet remodel (DWT) for multi-scale characteristic extraction. It is carried out thru Python and the PyWavelets library to offer an open-source, reproducible solution. The framework decomposes pictures into subscales of path and directional detail throughout multiple scales, extracting statistical and textural descriptors optimized for remote sensing obligations. A complete assessment of 500 multispectral patches (Sentinel-2, Landsat-8, and high-decision sensors) demonstrates advanced overall performance in land cover class, accomplishing an accuracy of 92.4%, outperforming uncooked pixel methods (84.1%), important issue evaluation (PCA) (87.3%), and GLCM-based totally techniques (89.6%). A sensitivity analysis famous that Daubeches wavelet 4 at decomposition level three improves function discriminability, in particular for agricultural textures (91.2% accuracy) and concrete limitations (IoU=0.873), while directional subbands (LH/HL) reduce transition area mistakes by way of 23%. The computational efficiency (184 ms/megapixel) remains possible. These consequences show that DWT is an effective and handy device for improving faraway sensing analysis, with the full code and datasets being made publicly available to promote community adoption and foster innovation.
Read MoreDoi: https://doi.org/10.54216/FPA.210110
Vol. 21 Issue. 1 PP. 142-154, (2026)
The task of automatically generating descriptive and accurate image tags has gained significant attention in recent years due to the exponential growth of image data. Traditional methods for image tagging rely on manual annotation, which is time-consuming and subjective. Automated imagine description fills the gap between visual content and human comprehension, making it vital for activities such as information retrieval, editing, and accessibility. The expanding number of unannotated photographs makes manual tagging impossible. This paper provides a deep learning-based system that combines CNNs for feature extraction, RNNs for caption production, and attention techniques to focus on significant image areas. The model uses a sequence-to-sequence architecture to create coherent captions using pre-trained CNN features and attention-enhanced RNNs. Experiments on datasets such as Flickr8k and Flickr30k show higher performance, as evidenced by BLEU, ROUGE, and CIDEr measures. This approach provides a scalable, cutting-edge solution for image captioning, with potential applications in video analysis, enriched language production, and larger datasets.
Read MoreDoi: https://doi.org/10.54216/FPA.210112
Vol. 21 Issue. 1 PP. 165-175, (2026)
Plant disease detection using deep learning has achieved high accuracy, but traditional centralized training poses significant privacy risks and incurs high data transmission costs. This study presents a privacy-preserving federated learning (FL) framework for plant disease diagnosis that enables decentralized model training across geographically distributed agricultural sites. Rather than transferring raw farm data to a central server, local models are trained on edge devices and share only model updates. To address data heterogeneity from diverse climates, soils, and plant species, we introduce adaptive aggregation strategies that improve model generalization. Furthermore, we incorporate differential privacy and homomorphic encryption to ensure secure model updates and protect sensitive information from potential breaches. Experimental evaluations on benchmark datasets, including Plant Village and real-world field images, show that the proposed FL-based system achieves comparable accuracy to centralized models while significantly enhancing data privacy and reducing communication overhead. The framework maintains over 93% classification accuracy across 38 plant disease categories, with minimal degradation from added privacy mechanisms. Additionally, we analyze the trade-off between accuracy and communication efficiency, demonstrating the method’s practicality in bandwidth-constrained rural environments. The proposed system offers a scalable, secure, and field-deployable solution for real-time plant disease monitoring, supporting the widespread adoption of AI in precision agriculture without compromising data confidentiality.
Read MoreDoi: https://doi.org/10.54216/FPA.210113
Vol. 21 Issue. 1 PP. 176-185, (2026)
Breast cancer is the most prevalent cancer-affecting women worldwide and remains a major cause of mortality. Early detection and accurate prognosis are critical to improving survival outcomes. This study introduces a novel predictive model for breast cancer diagnosis that integrates a lazy learning paradigm with the K-Nearest Neighbors (KNN) algorithm, optimized through a Linear Nearest Neighbor (NN) Search technique and the use of Euclidean distance as the similarity measure. The dataset, comprising 4,024 patient records with 15 clinical and demographic attributes, was obtained from a public repository and underwent rigorous preprocessing, including handling of missing values, normalization, and categorical encoding. The classification model was trained and evaluated using 1:9 cross-validation, with K values ranging from 1 to 9 and a constant batch size of 100 to identify the optimal configuration. Among various configurations tested, the model with K=5 demonstrated the highest performance, achieving an accuracy of 88.02%, precision of 0.87, and recall of 0.88. Additional performance metrics such as F-measure, Matthews Correlation Coefficient (MCC), and Kappa statistic further confirmed the robustness of the selected configuration. The proposed model shows superior predictive capability compared to traditional settings and can serve as a decision-support tool for clinicians. The findings suggest that the combination of lazy learning, effective neighbor search strategy, and robust distance metric can substantially enhance the predictive accuracy of breast cancer diagnosis. This study highlights the potential of machine learning-based tools in clinical oncology, offering a data-driven approach for early intervention and patient outcome improvement.
Read MoreDoi: https://doi.org/10.54216/FPA.210114
Vol. 21 Issue. 1 PP. 186-200, (2026)
Technological developments have produced methods that can generate educational videos from input text or sound. Recently, the use of deep learning techniques for image and video generation has been widely explored, particularly in education. However, generating video content from conditional inputs such as text or speech remains a challenging area. In this paper, we introduce a novel method to the educational structure, Generative Adversarial Network (GAN), which develop frame-for-frame frameworks and are able to create full educational videos. The proposed system is structured into three main phases in the first phase; the input (either text or speech) is transcribed using speech recognition. In the second phase, key terms are extracted and relevant images are generated using advanced models such as CLIP and diffusion models to enhance visual quality and semantic alignment. In the final phase, the generated images are synthesized into a video format, integrated with either pre-recorded or synthesized sound, resulting in a fully interactive educational video. The proposed system is compared with other systems such as TGAN, MoCoGAN, and TGANS-C, achieving a Fréchet Inception Distance (FID) score of 28.75%, which indicates improved visual quality and better over existing methods.
Read MoreDoi: https://doi.org/10.54216/FPA.210115
Vol. 21 Issue. 1 PP. 201-213, (2026)