Enhancing the Random Forest Model via Synthetic Minority Oversampling Technique for Credit-Card Fraud Detection

Journal of Computing Theories and Applications, Journal Year: 2024, Volume and Issue: 1(4), P. 407 - 420

Published: March 26, 2024

Fraudsters increasingly exploit unauthorized credit card information for financial gain, targeting un-suspecting users, especially as institutions expand their services to semi-urban and rural areas. This, in turn, has continued ripple across society, causing huge losses lowering user trust implications all cardholders. Thus, banks cum are today poised implement fraud detection schemes. Five algorithms were trained with without the application of Synthetic Minority Over-sampling Technique (SMOTE) assess performance. These included Random Forest (RF), K-Nearest Neighbors (KNN), Naïve Bayes (NB), Support Vector Machines (SVM), Logistic Regression (LR). The methodology was implemented tested through an API using Flask Streamlit Python. Before applying SMOTE, RF classifier outperformed others accuracy 0.9802, while accuracies LR, KNN, NB, SVM 0.9219, 0.9435, 0.9508, 0.9008, respectively. Conversely, after achieved a prediction 0.9919, whereas attained 0.9805, 0.9210, 0.9125, 0.8145, results highlight effectiveness combining SMOTE enhance detection.

Language: Английский

---

MFEUsLNet: Skin cancer detection and classification using integrated AI with multilevel feature extraction-based unsupervised learning

Engineering Science and Technology an International Journal, Journal Year: 2024, Volume and Issue: 51, P. 101632 - 101632

Published: Feb. 7, 2024

Skin Cancer is the most common form of disease and responsible for millions deaths each year. Most relevant studies concentrate on algorithms that are based machine learning, few deep learning as well. However, due to several challenges in dermoscopic image acquisition, these unable deliver highest possible level accuracy specificity. Therefore, this article implements skin cancer detection classification (SCDC) system using multilevel feature extraction (MFE)-based artificial intelligence (AI) with unsupervised (USL), here after denoted MFEUsLNet. Initially, given images preprocessed bilateral filter, which removes noise artifacts from source images. Then, a well-known USL approach named K-means clustering (KMC) used segmentation lesion, can detect affected lesion quite efficiently. gray co-occurrence matrix (GLCM), redundant discrete wavelet transform (RDWT) low level, texture colour extraction. Finally, recurrent neural network (RNN) classifier train multi-level features classify multiple types cancer. The simulations proven proposed MFEUsLNet model outperformed state-of-the-art SCDC approaches terms medical statistical quality metrics such accuracy, specificity, precision, recall, F1-score, sensitivity ISIC-2020 dataset.

Language: Английский

---

BEHeDaS: A Blockchain Electronic Health Data System for Secure Medical Records Exchange

Journal of Computing Theories and Applications, Journal Year: 2024, Volume and Issue: 1(3), P. 231 - 242

Published: Jan. 6, 2024

Blockchain platforms propagate into every facet, including managing medical services with professional and patient-centered applications. With its sensitive nature, record privacy has become imminent for patient diagnosis treatments. The nature of records continued to necessitate their availability, reachability, accessibility, security, mobility, confidentiality. Challenges these include authorized transfer on referral, security across platforms, content diversity, platform interoperability, etc. These, are today – demystified blockchain-based apps, which proffers platform/application achieve data features associated the records. We use a permissioned-blockchain healthcare management. Our choice permission mode hyper-fabric ledger that uses world-state peer-to-peer chain is smart contracts do not require complex algorithm yield controlled transparency users. Its actors patients, practitioners, health-related officers as users create, retrieve, store aid interoperability. population 500, system yields transaction (query https) response time 0.56 seconds 0.42 seconds, respectively. To cater scalability yielded 0.78 063 respectively, 2500

Language: Английский

---

Handling Transactional Data Features via Associative Rule Mining for Mobile Online Shopping Platforms

International Journal of Advanced Computer Science and Applications, Journal Year: 2024, Volume and Issue: 15(3)

Published: Jan. 1, 2024

Transactional data processing is often a reflection of consumer's buying behavior. The relational records if properly mined, helps business managers and owners to improve their sales volume. Transaction datasets are rippled with the inherent challenges in manipulation, storage handling due infinite length, evolution product features, concept, oftentimes, complete drift away from feat. previous studies' inability resolve many these as abovementioned, alongside assumptions that transactional presumed be stationary when using association rules – have been found also hinder performance. As it deprives decision support system needed flexibility robust adaptiveness manage dynamics concept characterizes transaction data. Our study proposes an associative rule mining model four consumer theories RapidMiner Hadoop Tableau analytic tools handle such large dataset was retrieved Roban Store Asaba consists 556,000 records. 6-layered framework yields its best result 0.1 value for both confidence level(s) at 94% accuracy, 87% sensitivity, 32% specificity, 20-second convergence time.

Language: Английский

---

An Interpretable Machine Learning Strategy for Antimalarial Drug Discovery with LightGBM and SHAP

Journal of Future Artificial Intelligence and Technologies, Journal Year: 2024, Volume and Issue: 1(2), P. 84 - 95

Published: Aug. 7, 2024

Malaria continues to pose a significant global health threat, and the emergence of drug-resistant malaria exacerbates challenge, underscoring urgent need for new antimalarial drugs. While several machine learning algorithms have been applied quantitative structure-activity relationship (QSAR) modeling compounds, there remains more interpretable models that can provide insights into underlying mechanisms drug action, facilitating rational design compounds. This study develops QSAR model using Light Gradient Boosting Machine (LightGBM). The is integrated with SHapley Additive exPlanations (SHAP) enhance interpretability. LightGBM demonstrated superior performance in predicting activity, an ac-curacy 86%, precision 85%, sensitivity 81%, specificity 89%, F1-score 83%. SHAP analysis identified key molecular descriptors such as maxdO GATS2m contributors activity. integration not only enhances predictive but also provides valuable importance features, aiding approach bridges gap between accuracy interpretability, offering robust framework efficient effective discovery against strains.

Language: Английский

---

RICE DISEASE RECOGNITION USING TRANSFER LEARNING XCEPTION CONVOLUTIONAL NEURAL NETWORK

Jurnal Teknik Informatika (Jutif), Journal Year: 2023, Volume and Issue: 4(6), P. 1535 - 1540

Published: Dec. 26, 2023

As one of the major rice producers, Indonesia faces significant challenges related to plant diseases such as blast, brown spot, tugro, leaf smut, and blight. These threaten food security result in economic losses, underscoring importance early detection management diseases. Convolutional Neural Network (CNN) has proven effective detecting plants. Specifically, transfer learning with CNN, particularly Xception model, advantage efficiently extracting automatic features performing well even limited datasets. This study aims develop model for disease recognition based on images. Through fine-tuning process, achieved accuracies, precisions, recalls, F1-scores 0.89, 0.90, respectively, a dataset total 320 Additionally, outperformed VGG16, MobileNetV2, EfficientNetV2.

Language: Английский

---

UNMASKING FRAUDSTERS: Ensemble Features Selection to Enhance Random Forest Fraud Detection

Journal of Computing Theories and Applications, Journal Year: 2023, Volume and Issue: 1(2), P. 201 - 211

Published: Dec. 25, 2023

Fraud detection is used in various industries, including banking institutes, finance, insurance, government agencies, etc. Recent increases the number of fraud attempts make crucial for safeguarding financial information that confidential or personal. Many types problems exist, card-not-present fraud, fake Marchant, counterfeit checks, stolen credit cards, and others. An ensemble feature selection technique based on Recursive elimination (RFE), Information gain (IG), Chi-Squared (X2) concurrence with Random Forest algorithm, was proposed to give research findings results prevention. The objective choose essential features training model. Receiver Operating Characteristic (ROC) Score, Accuracy, F1 Precision are evaluate model's performance. show model can differentiate between fraudulent transactions those not, an ROC Score 95.83% Accuracy 99.6%. 99.6%% precision 100% further sustain ability detect least false positives correctly. reduced time did not compromise performance, making it a valuable tool businesses preventing transactions.

Language: Английский

---

Application of Improved Lightweight Network and Choquet Fuzzy Ensemble Technology for Soybean Disease Identification

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 25146 - 25163

Published: Jan. 1, 2024

The identification of soybean disease images in natural scenes has been a challenging task due to their complex backgrounds and diverse spot patterns. Traditional single convolutional neural network (CNN) for image recognition often cannot have both high accuracy strong generalization ability. Therefore, this paper focuses on the classification leaf diseases using improved lightweight networks transfer learning, improves precision by introducing Choquet fuzzy ensemble strategy. First, long short-term memory (ConvLSTM) layer squeeze excitation (SE) block are introduced into four original models (Xception, MobileNetV2, NASNetMobile, MobileNet) improve network's ability grasp features, then confidence scores obtained from fed iensemble complete aggregation final results. In order performance model enrich distribution samples high-dimensional feature space, converts healthy diseased an unsupervised translation method based Cycle-Consistent Adversarial Networks (CycleGAN). results show that higher than network. proposed obtains 94.27% average F1-score 94% task, which is better other methods. It good application prospect initially meets production requirements identification.

Language: Английский

---

Metaheuristic Feature Selection for Diabetes Prediction with P-G-S Approach

Procedia Computer Science, Journal Year: 2025, Volume and Issue: 252, P. 165 - 171

Published: Jan. 1, 2025

Language: Английский

---

Classification of Diabetes According to Medical Indicators Using Machine Learning

Communications in computer and information science, Journal Year: 2025, Volume and Issue: unknown, P. 187 - 197

Published: Jan. 1, 2025

Language: Английский

---