JITK (Jurnal Ilmu Pengetahuan dan Teknologi Komputer)

OPTIMIZATION OF SVM ALGORITHM FOR OBESITY CLASSIFICATION WITH SMOTE TECHNIQUE AND HYPERPARAMETER TUNING

2025-08-01T07:47:07+00:00

Excessive fat accumulation that impairs personal health and raises the risk of chronic diseases is the hallmark of obesity, a global health issue. Decision Tree (DT) has been widely used for obesity classification, but it tends to suffer from overfitting and poor performance on imbalanced datasets. To overcome these limitations, this study proposes an optimization of the Support Vector Machine (SVM) algorithm using Synthetic Minority Over-sampling Technique (SMOTE) and Hyperparameter Tuning. SMOTE was applied to balance the class distribution, whereas Grid Search was utilized to determine the optimal combination of hyperparameters (C, gamma, and kernel). The dataset employed in this research comprises multiple features related to individual health and lifestyle, with obesity level as the target class. The experimental results demonstrate that the optimized SVM model demonstrated strong classification performance, attaining 97% in accuracy, precision, recall, and F1-score. This high performance is significant because it enables more accurate early detection of obesity risk, which can support timely medical intervention and personalized treatment planning, ultimately contributing to better public health outcomesThese findings suggest that incorporating SMOTE and Hyperparameter Tuning substantially improves SVM performance, establishing it as a robust approach for obesity classification and early risk detection.

SYSTEMATIC LITERATURE REVIEW ON ARTIFICIAL INTELLIGENCE IN INDONESIA’S PUBLIC SECTOR: REIMAGINING DIGITAL GOVERNMENT

2025-08-19T04:42:51+00:00

This study conducts a Systematic Literature Review (SLR) to critically examine the application of Artificial Intelligence (AI) in e-government within the Indonesian public sector. Addressing the limited empirical research and fragmented understanding of AI adoption in Indonesia’s digital governance landscape, this review analyzes 22 peer reviewed articles published between 2021 and 2025 from reputable databases including Scopus, IEEE, ACM Digital Library, SpringerLink, and Emerald Insight. The review identifies adaptability and innovation, ethical consideration, collaboration and partnership as the most frequently cited critical success factors. Meanwhile, the top three recurring challenges are lack of awareness, skill & expertise, policy or legal uncertainty, resistance to change. To address these challenges, the study proposes a multi dimensional AI implementation strategy focusing on strengthening digital infrastructure, developing human capital through sustained capacity building, formulating clear and accountable AI governance policies, and fostering inclusive, cross sectoral stakeholder engagement. This study offers novel insights by mapping AI related factors into the Technology,Organization, Environment (TOE) framework and synthesizing practical, context-specific recommendations for Indonesian policymakers seeking to build an adaptive, inclusive, and sustainable AI based e-government ecosystem

PERFORMANCE EVALUATION OF NEWTON–KONTOROVICH AND ADAPTIVE NEWTON LINE SEARCH ON MULTIVARIATE NONLINEAR SYSTEMS

2025-10-07T05:56:25+00:00

Solving multivariate nonlinear systems is essential in engineering, physics, and applied sciences. This study compares the performance of two numerical methods—Newton–Kontorovich and Interactive Newton–Raphson with Line Search—on trigonometric and exponential nonlinear systems. The methods are evaluated based on convergence rate, accuracy, and iteration efficiency through numerical simulations using MATLAB. The Newton–Kontorovich method, typically used for integral or differential equations, is compared with the adaptive line search strategy that enhances global convergence. Results show that the Interactive Newton–Raphson method achieves a smaller final error (5.95×10⁻²) with stable convergence, while Newton–Kontorovich converges in fewer iterations but with larger error (3.126). These findings highlight the superiority of adaptive strategies for complex nonlinear systems. Practical implications include improved numerical reliability for applications in structural engineering, optimization, and scientific modeling.

SENTIMENT ANALYSIS OF IT WORKERS ON NO CODE AND LOW CODE TRENDS: COMPARISON OF LSTM AND SVM MODELS

2025-09-17T06:56:58+00:00

This research explores the sentiment of IT professionals toward the growing trend of No Code and Low Code technologies by comparing the performance of Support Vector Machine (SVM) and Long Short-Term Memory (LSTM) algorithms. Using the SEMMA methodology and automatic labeling with ChatGPT, a total of 4,238 comments were collected from Reddit and Twitter and categorized into positive, neutral, and negative sentiments. The analysis showed that neutral sentiment dominates on both platforms (47.9% on Reddit and 48.8% on Twitter), followed by positive sentiment (41.3% and 43.1%, respectively), indicating cautious but optimistic attitudes toward LCDPs. In terms of model performance, SVM outperformed LSTM with 87% accuracy and a weighted F1-score of 0.87, compared to LSTM’s 80% accuracy and a weighted F1-score of 0.80. These findings confirm that classical machine learning methods remain highly effective for short-text sentiment analysis in social media, particularly when combined with TF-IDF feature representation, SMOTE balancing, and LLM-based automatic labeling, while also offering new insights into IT community perceptions of disruptive technologies

IMPROVING HANDWRITTEN DIGIT RECOGNITION USING CYCLEGAN-AUGMENTED DATA WITH CNN–BILSTM HYBRID MODEL

2025-09-17T08:48:29+00:00

Handwritten digit recognition presents persistent challenges in computer vision due to the high variability in human handwriting styles, which necessitates robust generalization in classification models. This study proposes an advanced data augmentation strategy using Cycle-Consistent Generative Adversarial Networks (CycleGAN) to improve recognition accuracy on the MNIST dataset. Two architectures are evaluated: a standard Convolutional Neural Network (CNN) and a hybrid model combining CNN for spatial feature extraction and Bidirectional Long Short-Term Memory (BiLSTM) for sequential pattern modeling. The CycleGAN-based augmentation generates realistic synthetic images that enrich the training data distribution. Experimental results demonstrate that both models benefit from the augmentation, with the CNN-BiLSTM model achieving the highest accuracy of 99.22%, outperforming the CNN model’s 99.01%. The study’s novelty lies in the integration of CycleGAN-generated data with a CNN–BiLSTM architecture, which has been rarely explored in previous works. These findings contribute to the development of more generalized and accurate deep learning models for handwritten digit classification and similar pattern recognition tasks.

OPTIMIZATION OF EFFICIENTNET-B0 ARCHITECTURE TO IMPROVE THE ACCURACY OF GLAUCOMA DISEASE CLASSIFICATION

2025-10-13T09:20:25+00:00

Glaucoma is a chronic eye disease that can potentially cause permanent blindness if not detected early. This study aims to improve the generalization capability and reliability of glaucoma classification by optimizing the EfficientNetB0 architecture based on a Convolutional Neural Network (CNN). Optimization was carried out by applying double dropout (0.4 and 0.3) and adding a Dense layer with 128 ReLU-activated neurons to reduce overfitting and strengthen non-linear feature representation. The dataset used consists of 1,450 fundus images (899 glaucoma and 551 normal) obtained from IEEE DataPort. Model performance evaluation was performed using accuracy, precision, recall (sensitivity), specificity, F1 score, and Area Under the Curve (AUC) metrics, complemented by confusion matrix analysis to assess overall classification performance. The results showed that the optimized EfficientNetB0 model consistently outperformed the baseline comparison model with the highest accuracy, precision, recall (sensitivity), specificity, F1 score, and AUC values of 95%. Based on the system performance results obtained, the Proposed model can be used as an aid for medical personnel in classifying glaucoma conditions so that they can provide appropriate medical treatment and reduce the risk of permanent blindness due to glaucoma.

COMPARISON OF PRINCIPAL COMPONENT ANALYSIS AND RANDOM FOREST ALGORITHM FOR PREDICTING HOUSING PRICES

2025-10-07T08:04:13+00:00

House price predictions are an important thing in the property industry and are useful for buyers in making decisions. Principal Component Analysis (PCA) and Random Forest (RF) methods were used for accuracy analysis in predicting housing prices. Purpose of this research is to measure the accuracy of both methods also to compare RF method optimized with PCA and the one that has not been optimized. The data used is house prices in Karanganyar city based on data scraping results on the rumah123.com site. The analysis reveals that Jaten has the highest number of house sales, and sales of houses with land ownership certificates are also the highest. Of the 10 variables used, land area and buildings have the most influence on selling prices. The model training results show that the RF and PCA methods combination has more optimal value than only using the RF method. The error rate of the PCA method is smaller, averaging 0.0257, making its value more consistent than using only the RF method, which has a larger error value with an average of 0.0332. The model training time using PCA is faster (5005.75) than only using the RF method (6099.25)

APPLICATION OF RANDOM FOREST ALGORITHM FOR ARRHYTHMIA DETECTION BASED ON ELECTROCARDIOGRAM DATA

2025-10-14T09:43:11+00:00

Arrhythmia is a common cardiac disorder that requires early detection to prevent serious complications. This study applied the Random Forest algorithm to enhance electrocardiogram (ECG) analysis and enable accurate arrhythmia classification. Unlike prior studies that focused primarily on resting ECG signals, this research incorporated dynamic data collected from 26 participants performing three physical activities for three minutes each, capturing physiological variations across multiple activity states. The Random Forest model was constructed and evaluated using ECG-derived temporal and morphological features to detect potential arrhythmias. Experimental results showed that the model achieved an accuracy of 97.4%, with precision, recall, and F1-score each reaching 98%, and an AUC of 0.97. However, several limitations remain, including the relatively small and homogeneous sample, as well as the short recording duration. Nonetheless, the proposed approach demonstrates strong potential to support early cardiac screening and real-time monitoring, particularly in portable and resource-limited healthcare applications

COMPARATIVE STUDY OF TRANSFORMER-BASED MODELS FOR AUTOMATED RESUME CLASSIFICATION

2025-10-27T07:38:09+00:00

This study presents a comparative evaluation of transformer-based models and traditional machine learning approaches for automated resume classification—a key task in optimizing recruitment workflows. While traditional approaches like Support Vector Machines (SVM) with TF-IDF demonstrated the highest performance (93.26% accuracy and 95% F1-score), transformer models such as DistilBERT and RoBERTa showed competitive results with 93.27% and 91.34% accuracy, respectively, and fine-tuned BERT achieved 84.35% accuracy and an F1-score of 81.54%, indicating strong semantic understanding. In contrast, Word2Vec + LSTM performed poorly across all metrics, highlighting limitations in sequential modelling for resume data. The models were evaluated on a curated resume dataset available in both text and PDF formats using accuracy, precision, recall, and F1-score, with preprocessing steps including tokenization, stop-word removal, and lemmatization. To address class imbalance, we applied stratified sampling, macro-averaged evaluation metrics, early stopping, and simple data augmentation for underrepresented categories. Model training was conducted in a PyTorch environment using Hugging Face’s Transformers library. These findings highlight the continued relevance of traditional models in specific NLP tasks and underscore the importance of model selection based on task complexity and data characteristics

WORD2VEC OPTIMALIZATION USING TRANSFER LEARNING IN INDONESIAN LANGUAGE FOR HIGHER EDUCATION

2025-10-16T03:16:05+00:00

Natural language processing (NLP) in Indonesian faces challenges due to limited linguistic resources, particularly in developing optimal word embedding models. This study optimizes the Word2Vec model for Indonesian in higher education contexts by leveraging transfer learning and lexicon expansion. Using a dataset of 4,463 higher education related tweets consisting of positive and negative sentiment categories, the proposed NewWord2Vec model combined with a Support Vector Machine (SVM) classifier achieved a 4% improvement in word detection accuracy compared to the standard Word2Vec. This enhancement demonstrates better performance in capturing linguistic nuances and sentiment orientation in Indonesian text. However, the model’s applicability remains limited to higher education terminology, and potential biases from transfer learning must be addressed. Future research should expand the dataset to diverse domains and refine the transfer learning process to better capture contextual variations in Indonesian. These findings contribute to advancing NLP applications in Indonesian, particularly for automated assessment systems, recommendation tools, and academic decision-making processes

INTEGRATED SYSTEM-BASED SMART APPLICATION (SIPATIN) FOR STRENGTHENING FISHERIES GOVERNANCE IN LEBAK REGENCY, BANTEN

2025-10-06T08:02:00+00:00

The Fisheries Office of Lebak Regency is responsible for the management of capture fisheries, aquaculture, resource monitoring, and the marketing of fishery products. However, geographical challenges, the difficulty of obtaining real-time data, and the use of conventional monitoring and reporting methods hinder effective and sustainable fisheries governance. In addition, limited market reach—primarily targeting only local areas—further restricts the region’s economic potential. This study aims to address issues related to monitoring, reporting, and marketing through the development of an Integrated Fisheries Information System (SIPATIN), a smart mobile-based fisheries governance application integrated with a web-based monitoring platform. SIPATIN features include fishery area mapping, real-time reporting, an E-Commerce marketing platform, and a recommendation system that provides detailed information on products, seller locations, prices, reviews, estimated delivery times, and proximity-based suggestions for users. The system was developed using a prototyping method, consisting of needs analysis, design, development, testing, and evaluation based on user feedback. The application was evaluated using the System Usability Scale (SUS), which scored 74.26 (Good), and User Acceptance Testing (UAT), involving 246 respondents and resulting in a score of 79.13 (Acceptable). The results of this study show that SIPATIN effectively supports integrated fisheries governance, enhances service efficiency at the Lebak Regency Fisheries Office, and empowers fishery business actors including fishers, fish farmers, and small and medium enterprises (SMEs) in processed fish products. Furthermore, this research also produces a data-based fishery system for sustainable economic development

RE-DESIGNING JAKLINGKO APPS UI/UX USING AGILE REQUIREMENT ENGINEERING APPROACH

2025-09-17T07:30:29+00:00

Public transportation has become a staple in a lot of countries, including Indonesia. As the largest city in Indonesia, is trying to accommodate the dense traffic in Jakarta by implementing various types of public transportation, one of which is the Bus Rapid Transit (BRT). BRT has its own application called Jaklingko, which the commuter uses to gain information about the BRT. Unfortunately, this application has bad reviews in the app store. This research tried to redesign the UI/UX of this application using prototyping and the System Usability Scale (SUS) as tools for agile requirement engineering tools. In Agile requirements usually conducted the same as traditional which is using interview or observation. But, using this method proved to be time consuming. Therefore this research tried to incorporate prototyping and SUS into the requirements gathering process. After the requirements are collected, the next phase is redesigning the application based on the gathered requirements. From the research conducted, the main pain point of the responses is how much information is given in the apps. This research also found that prototyping and SUS could be used to gather requirements, but they will depend heavily on the test case being used. Therefore, it is not suitable for stand alone gathering tools but good as a confirmation tool

REFORMULATION OF MULTI-ATTRIBUTE UTILITY THEORY NORMALIZATION TO HANDLE ASYMMETRIC DATA IN MADM

2025-10-06T08:57:39+00:00

Multi-Attribute Utility Theory (MAUT) is a widely used multi-attribute decision-making (MADM) method due to its ability to integrate multiple criteria into a single utility value. However, conventional MAUT faces limitations when handling asymmetric data, where standard normalization processes often lead to value distortion and less representative rankings. This study aims to reformulate the normalization function in MAUT to improve adaptability to non-symmetric data distributions and to enhance ranking validity in decision-making. A modification approach called MAUT-A was developed by applying an adaptive normalization mechanism capable of accommodating extreme distributions and outliers by adding Z-score normalization. The performance of MAUT-A was evaluated by comparing the correlation of its ranking results with reference rankings, and the outcomes were benchmarked against conventional MAUT. The experimental findings indicate that conventional MAUT achieved a correlation value of 0.9688 with the reference ranking, while the proposed MAUT-A method achieved a higher correlation of 0.9792. This improvement represents that MAUT-A has better suitability, stability, and reliability in managing asymmetric data. The study contributes by offering a reformulated MAUT framework through adaptive normalization, providing more accurate, stable, and fair ranking outcomes. This approach enhances the validity of MADM applications, particularly in contexts involving asymmetric data distributions

UNVEILING SPATIAL PATTERNS OF LAND CONVERSION THROUGH MACHINE LEARNING AND SPATIAL DISTRIBUTION ANALYSIS

2025-10-08T02:17:53+00:00

Kayu Agung District in Ogan Komering Ilir (OKI) Regency, South Sumatra, has undergone rapid population growth, resulting in notable land-use transformations. This study examines land-use change dynamics from 2019 to 2023 and identifies their spatial distribution using satellite imagery. Satellite imagery classification was performed using three machine learning algorithms—K-Nearest Neighbors (KNN), Naïve Bayes, and Logistic Regression—with KNN achieving the highest accuracy. Spatial analysis employing the Variance-to-Mean Ratio (VMR) revealed that land-use changes are spatially clustered, indicating concentrated land conversion in specific areas. These findings emphasize potential environmental risks, including declining green open spaces and increasing urban pressure. The study contributes by integrating machine learning and spatial statistical analysis (VMR) as a comprehensive framework for understanding land-use conversion, providing scientific insights to support adaptive spatial planning and the achievement of Sustainable Development Goal (SDG) 11: Sustainable Cities and Communities.

MONITORING ELDERLY HEART RATE BASED ON OXIMETER SENSORS

2025-09-17T07:09:16+00:00

Heart rate check is an important step in preventing heart attacks that is often not realized by the elderly. However, independent heart rate checks by the elderly have not utilized technology, especially Android. This study design a heart rate detector using the Max30102 Oximeter Sensor integrated with Android device from the elderly aged 60 to 75 years and displays the results of the heart rate per minute (BPM) along with normal or abnormal status on the Android application. The prototype method involves the stages of development, testing, and evaluation of the tool. The results of the study showed that this heart rate detector was able to provide data on heart rate per minute (BPM) that was accurate and easily accessible to the elderly, so that the elderly could check their health independently. The test results indicate a detection accuracy of 97% with a standard deviation of 1.19 BPM, which is higher compared to studies using the Max30100. Thus, this tool is expected to help increase the independence of the elderly in monitoring heart health and reduce the risk of heart attack through routine monitoring

OPTIMIZING SHUFFLENET WITH GRIDSEARCHCV FOR GEOSPATIAL DISASTER MAPPING IN INDONESIA

2025-10-28T02:16:32+00:00

Accurate classification of natural disasters is crucial for timely response and effective mitigation. However, conventional approaches often suffer from inefficiency and limited reliability, highlighting the need for automated deep learning solutions. This study proposes an optimized Convolutional Neural Network (CNN) based on the lightweight ShuffleNet architecture, enhanced through GridSearchCV for systematic hyperparameter tuning. Using a geospatial dataset of 3,667 images representing earthquake, flood, and wind-related disasters in Indonesia, the optimized ShuffleNet model achieved a peak accuracy of 99.97%, outperforming baseline CNNs such as MobileNet, GoogleNet, ResNet, DenseNet, and standard ShuffleNet. While these results demonstrate the potential of combining lightweight architectures with automated optimization, the exceptionally high performance also indicates possible risks of overfitting and dataset bias due to limited variability. Therefore, future research should validate this approach using larger, multi-source datasets to ensure robustness and real-world applicability

YOLO MODEL DETECTION OF STUDENT NEATNESS BASED ON DEEP LEARNING: A SYSTEMTIC LITERATURE REVIEW

2025-10-06T09:12:25+00:00

Maintaining proper student neatness (uniform compliance, grooming standards, and posture) is essential for fostering disciplined learning environments. While traditional monitoring methods are labor-intensive and subjective, computer vision-based solutions leveraging You Only Look Once (YOLO) architectures offer promising alternatives. The objective of this study is to evaluate YOLO optimization techniques for student neatness detection, identify key challenges, and propose relevant future research directions. This systematic review evaluates 28 recent studies (2021-2024) to analyze optimization techniques for YOLO models in student neatness detection applications. Key findings demonstrate that attention-enhanced variants (e.g., YOLOv10-MSAM) achieve 87.0% mAP@0.5, while pruning and quantization methods enable real-time processing (50-130 FPS) on edge devices like Jetson Orin. The analysis reveals three critical challenges: (1) occlusion handling in crowded classrooms (10-15% false negatives), (2) lighting/background variability, and (3) ethical concerns regarding facial recognition. Emerging solutions include hybrid vision-language models for explainable detection and federated learning for privacy preservation. The review proposes a taxonomy of optimization approaches categorizing architectural modifications (attention mechanisms, lightweight backbones), data augmentation strategies (GAN-based synthesis), and deployment techniques (TensorRT acceleration). Future research directions emphasize multi-modal sensor fusion and domain adaptation for cross-institutional generalization. This work provides educators and AI developers with evidence-based guidelines for implementing automated neatness monitoring systems while addressing practical constraints in educational settings.

COMPARATIVE ANALYSIS OF HYPERPARAMETER OPTIMIZATION TECHNIQUES ON LIGHTGBM FOR ASTHMA PREDICTION

2025-10-28T05:47:29+00:00

This study presents a comparative study of hyperparameter optimization methods applied to the Light Gradient Boosting Machine (LightGBM) algorithm for asthma prediction. Traditional machine learning models often face limitations in accuracy and generalization capabilities due to suboptimal hyperparameter configurations. To address these challenges, this study evaluates and compares four approaches: Default LightGBM, RandomizedSearchCV, Optuna Optimization, and Bayesian Optimization. Experimental results show that Bayesian Optimization provides the best performance with an accuracy of 78%, a precision of 0.7778, a recall of 0.7778, an F1-score of 0.7778, and an ROC-AUC of 0.975. These findings emphasize the importance of selecting an appropriate optimization strategy to improve model performance in clinical prediction tasks. Overall, this study confirms the effectiveness of Bayesian Optimization in improving the predictive capabilities of LightGBM and provides an important contribution to the development of decision support systems in healthcare, particularly in the diagnosis and management of asthma

DEVELOPMENT OF A SMART PARKING SYSTEM USING AUTOMATIC DEBIT AND OPTICAL CHARACTER RECOGNITION

2025-10-28T04:39:50+00:00

The current parking infrastructure predominantly relies on traditional or semi-automatic mechanisms, leading to significant inefficiencies during peak hours. This study proposes the development of a fully automated smart parking system utilizing locally sourced Indonesian components to reduce dependence on imported parts. The proposed Auto-Debit Smart Parking System incorporates Optical Character Recognition (OCR) for vehicle identification and automated payment, improving both accuracy and operational efficiency. The system consists of two primary modules: server software for gate control and an image-processing host application. Space Vector Pulse Width Modulation (SVPWM) is employed for switching control, and communication is facilitated via wired or wireless channels using the RS232C standard. Vehicle entry and exit are detected by sensors that transmit signals to the Command TX module. To evaluate real world applicability, the system was implemented and tested in various public and commercial environments, including office buildings, shopping malls, and open parking areas.These testing sites represent common urban parking conditions with varying lighting, network connectivity, and traffic density, allowing the system’s adaptability and reliability to be analyzed comprehensively. An experimental research method is adopted, encompassing prototype development, testing, data acquisition, and performance evaluation. The results indicate reduced operational costs and enhanced user convenience, validating the system’s effectiveness in supporting modern, efficient parking management

ENHANCING COFFEE PRODUCTION FACTOR ASSESSMENT USING LINEAR REGRESSION AND AHP FOR RELIABLE WEIGHT CONSISTENCY

2025-10-22T03:51:03+00:00

The agricultural sector, particularly coffee production, plays a crucial role in Indonesia’s economy as both a domestic commodity and an export product. However, efforts to optimize coffee production are often constrained by traditional Multi-Criteria Decision-Making (MCDM) methods that rely heavily on subjective judgments, leading to potential inconsistencies—especially in the presence of multicollinearity among variables. This study addresses that challenge by proposing a data-driven and consistent weighting method that integrates Multiple Linear Regression (MLR) with the Analytic Hierarchy Process (AHP). Regression coefficients derived from MLR—based on variables such as the area of immature (-0.2419), mature (0.8357), and damaged (0.5119) coffee plantations—are normalized and incorporated into the AHP pairwise comparison matrix. The resulting Consistency Ratio (CR) values are all below 0.1, indicating high internal consistency and statistical reliability of the derived weights. This integrated approach offers an objective and systematic foundation for MCDM in coffee production analysis, enhances the accuracy of agricultural planning, and supports evidence-based policymaking, while also providing a replicable model for addressing similar challenges in other sectors

COMPARATIVE ANALYSIS OF CLASSIFICATION ALGORITHMS IN HANDLING IMBALANCED DATA WITH SMOTE OVERSAMPLING APPROACH

2025-10-08T08:55:00+00:00

Most machine learning algorithms tend to yield optimal results when trained on datasets with balanced class proportions. However, their performance usually declines when applied to data with significant class imbalance. To address this issue, this study utilizes the Synthetic Minority Oversampling Technique (SMOTE) to improve class distribution before model training. Several classification algorithms were employed, including Decision Tree, K-Nearest Neighbors, Logistic Regression, Support Vector Machine, and Random Forest. Experimental results reveal that the Random Forest model produced the highest accuracy (95.70%) and the best F1-score, demonstrating a well-balanced trade-off between precision and recall. In contrast, the Logistic Regression algorithm achieved the highest recall (74.20%), indicating better sensitivity in identifying positive instances despite a lower F1-score. These outcomes highlight the importance of choosing appropriate classification methods based on the specific evaluation goals whether prioritizing accuracy, recall, or overall model balance.

PREDICTIVE MODEL FOR COOPERATIVE LOAN RECIPIENT ELIGIBILITY USING SUPERVISED MACHINE LEARNING

2025-10-28T01:58:39+00:00

Non-performing loans remain a critical challenge for cooperatives as they can undermine financial stability, erode member trust, and impede institutional growth. This study develops a predictive model for cooperative loan eligibility using supervised machine learning techniques and a novel three-class classification framework, Approved, Consideration, and Rejected, to support more objective and transparent decision-making. A dataset of 1,000 borrower records containing demographic and financial attributes was analyzed using Naive Bayes, Decision Tree, and Random Forest algorithms implemented in RapidMiner. The Random Forest algorithm achieved the best predictive performance with an accuracy of 96.02%, demonstrating its robustness and reliability compared to the other models. The proposed three-class system differentiates this study from conventional binary classification approaches, enabling finer distinctions among borrower categories and promoting fairness in cooperative credit evaluations. The findings provide practical guidance for cooperatives to adopt data-driven, transparent, and accountable decision-making systems that reduce manual bias and strengthen financial inclusion. Overall, the proposed three-class model built through a supervised learning framework offers a reliable, fair, and scalable solution to support sustainable lending practices and enhance risk management in cooperative institutions.

RTOS-BASED SYSTEM FOR TODDLER NUTRITIONAL STATUS DETECTION

2025-11-07T13:22:34+00:00

Determining the nutritional status of toddlers is essential for monitoring growth and preventing long-term health problems. Manual assessment requires significant time and is prone to human error; therefore, an automatic detection system based on height and weight parameters is needed. This study aims to develop a Real-Time Operating System (RTOS)–based system to detect the nutritional status of children aged 24–60 months, capable of managing task priorities, ensuring timely execution, and preventing race conditions using semaphores. The system employs an ultrasonic sensor to measure height, load cell sensors to measure body weight, and a web-based interface to input gender and age. Nutritional classification is determined through Z-score calculations using WHO reference data. Tests conducted on 200 children in various locations showed that the ultrasonic sensor achieved an average absolute error of 0.39 cm, a relative error of 0.409%, and an accuracy of 99.59%, while the load cell sensor achieved an average absolute error of 0.22 kg, a relative error of 1.587%, and an accuracy of 98.41%. The average execution times for the measurement and Z-score computation tasks were 4014.4 ms and 11.31 ms, respectively. The nutritional status classification results showed accuracy levels of 99.5% for Weight-for-Age (W/A), 99.5% for Height-for-Age (H/A), and 97.5% for Body Mass Index-for-Age (BMI/A) compared with manual assessments. The developed system demonstrated reliable performance in measurement and classification, with results consistent with conventional methods, indicating its potential as an efficient and accurate tool to assist healthcare workers in monitoring toddler nutrition status

QUANTUM-ASSISTED FEATURE SELECTION FOR IMPROVING PREDICTION MODEL ACCURACY ON LARGE AND IMBALANCED DATASETS

2025-10-14T10:12:58+00:00

One of the biggest obstacles to creating precise machine learning models is choosing representative and pertinent characteristics from big, unbalanced datasets. While too many features raise the risk of overfitting and computational expense, class imbalance frequently results in decreased accuracy and bias. The Simulated Annealing technique is used in this study to tackle a Quadratic Unconstrained Binary Optimization (QUBO) problem that is formulated as a quantum-assisted feature selection method to handle these problems. The technique seeks to reduce inter-feature redundancy and the number of selected features. There are 102,487 samples in the majority class and 11,239 in the minority class, totaling 28 characteristics in the experimental dataset. Nine ideal features were found during the feature selection method (12, 14, 15, 22, 23, 24, 25, 27, and 28). Ten-fold cross-validation was used to assess a Random Forest Classifier that was trained using an 80:20 split. With precision, recall, f1-score, and accuracy all hitting 1.00, the suggested QUBO+SMOTE method demonstrated exceptional performance. Comparatively, QUBO without SMOTE performed worse with accuracy 0.95 and minority-class f1-score of only 0.71, whereas a traditional Recursive Feature Elimination (RFE) approach obtained accuracy 0.97 with minority-class f1-score of 0.94. These findings indicate that QUBO can reduce dimensionality and address class imbalance which requires its integration with SMOTE. This study demonstrates how quantum computing can enhance the effectiveness and efficiency of machine learning, especially for large-scale imbalanced datasets

ECG-BASED ARRHYTHMIA DETECTION USING THE NARROW NEURAL NETWORK CLASSIFIER

2025-10-14T10:01:16+00:00

Electrocardiograms (ECG) are important for detecting arrhythmias. Conventional models such as CNN and LSTM are accurate but require large amounts of computation, making them difficult to use on wearable devices and for real-time monitoring. This study evaluates the Narrow Neural Network Classifier (NNNC) as a lightweight and efficient alternative. The dataset consists of 21 subjects with 881 ECG samples, categorized based on walking, sitting, and running activities, and processed through bandpass filtering, normalization, and P-QRS- T wave segmentation. The data is divided into training (70%), validation (15%), and test (15%) sets. The NNNC has 11 convolutional layers, a ReLU activation function, a Softmax output, and 120,000 parameters. The model was trained using the Adam optimizer, a batch size of 32, and a learning rate of 0.001 for 100 epochs and compared with SVM, CNN, and LSTM using accuracy, precision, recall, F1-score, and ROC-AUC. The results show that NNNC achieves an accuracy of 98.9%, a precision of 99.2%, a recall of 99.2%, and an F1-score of 99.2%, higher than SVM and comparable to CNN/LSTM, with lower computational consumption. The model is capable of reliably detecting early arrhythmias. These findings support the potential of NNNC for ECG-based automatic diagnostic systems, including real-time implementation on wearable devices, although further research is needed for large-scale validation