Jul 2026· 2026 4th International Conference on Sustainable Computing and Smart Systems (ICSCSS)· pp. 918-925· 0 citations· 14 references
Abstract
In this paper, a forensic end-to-end machine learning pipeline to detect financial fraud in PaySim synthetic mobile money dataset a corpus of 6,362,620 transactions with an extreme class imbalance ratio of 0.129% is presented. Proposed system is based on the distributed MLlib framework of Apache Spark in order to overcome the scalability issues associated with processing millions of financial records. Multicollinear raw ledger features are substituted with seven theoretically-based feature engineering transformations such as accounting discrepancy indices and account drain ratios as well as high-risk categorical flags. Ten classifiers such as Logistic Regression, Decision Tree variants, Random Forest ensembles, Gradient Boosted Trees, Linear SVC and a Factorization Machine are considered in a forensic model tournament. The imbalance between classes is alleviated through random undersampling to a balanced 50:50 training sample. The Matthews Correlation Coefficient (MCC) is the statistically rigorous evaluation index of the imbalanced binary classification that allows model selection. The winning ensemble model has a Recall of 99.39%, a Precision of 100.00%, an MCC of 0.9941 and a Financial Protection Index of over 96% which shows the operational reliability and good business value in preserving capital. Limitations regarding the synthetic nature of the data, the absence of sequence modeling, the schema dependence of the engineered features and the batch nature of the deployment are also discussed.
The study introduces a Stacked Logistic Regression ensemble to combine the predictive capacity of optimized Random Forest and XGBoost base classifiers and reveals that the proposed stacked model performance surpasses both individual base models.
Uduh Israel Akakoh, G. N. Edegbe· FUDMA Journal of Sciences· 0 citations
Credit card fraud remains a major challenge for financial institutions, both financially and operationally, as digital transactions continue to grow and fraud datasets remain highly imbalanced. This study compares the performance of several supervised machine learning models for fraud detection, using a unified data preprocessing pipeline. The approach includes removing duplicates, applying RobustScaler normalization, engineering features and using the Synthetic Minority Oversampling Technique (SMOTE) to balance classes before training. Four models were developed and tested Logistic Regression, Decision Tree, Random Forest and Artificial Neural Network (ANN) using the publicly available Kaggle Credit Card Fraud Detection dataset. Their performance was measured with Accuracy, Precision, Recall, F1-score and ROC-AUC metrics. Results showed that thorough preprocessing combined with SMOTE significantly improved the models ability to detect fraudulent transactions. Among them, the Random Forest model delivered the strongest overall performance, proving especially effective at handling highly imbalanced financial data. The comparative analysis also highlighted that ensemble learning methods generally outperform single classifiers in both accuracy and minority-class recognition. These findings emphasize the importance of pairing robust preprocessing strategies with machine learning techniques to boost fraud detection in real-world financial systems. The proposed system offers institutions a scalable and practical solution for building intelligent fraud detection systems, while laying the groundwork for future integration of Explainable AI (XAI) and real-time detection tools.
Nafiu Yahuza, Ahmad Baita Garko, Abubakar Atiku Muslim et al.· Lead Sci Journal of Manageme...· 0 citations
Digital payment services now handle millions of transactions each day, where even a tiny
fraction of fraud causes major financial losses and undermines customer trust. This paper
investigates how to accurately detect fraudulent transactions in a Kaggle financial payment
services dataset and to understand which transaction behaviors make payments appear risky.
The original dataset is pre-processed through outlier removal, label encoding, standardisation
and two stages balancing strategy combining random down sampling of the majority class with
SMOTE oversampling. Nine numeric and categorical features are then used as inputs to
classifiers including LR, KNN, DT, RF, SVM, GNB, AdaBoost, Bagging, Voting and Stacking
under four settings. They are no dimensionality reduction (NoDR), UMAP, NCA and PLS-DA
respectively. Hyperparameters are tuned with GridSearchCV using both 70/30 train–test split
and 10-fold cross-validation. Model performance is evaluated with accuracy, precision, recall,
F1-score, specificity, ROC-AUC, PR-AUC and training time. SHAP is applied to interpret
feature importance and local decisions. Without dimensionality reduction Bagging, Random
Forest and Stacking achieve 99.3 to 99.4% accuracy and F1, with ROC-AUC and PR-AUC close
to 99.95% on the test set and similarly strong cross-validation scores. UMAP and NCA preserve
high performance in lower dimensional spaces. While PLS-DA gives moderate but consistent
results. The study shows that ensemble models on the original feature space provide highly
accurate and stable fraud detection on this dataset. SHAP analysis reveals that source and
destination balances, transaction amount and type are the most influential features.
Merit Chinonso Opara· IIARD INTERNATIONAL JOURNAL...· 0 citations
Three deep tabular models, namely, an advanced multilayer perceptron (AdvancedMLP), an attention‐based residual network (AttentionFraudNet), and an advanced residual network (AdvancedResNet), are compared against three traditional machine learning baselines, including Random Forest, Gradient Boosting, and Logistic Regression.
Vahid Azarvand, Parvin Azhdari, A. Beitollahi· Engineering Reports· 0 citations
Credit card fraud is a serious threat in the digital financial ecosystem and is characterised by extreme class imbalance, with fraudulent transactions typically below 1%. This study compares two standard classification algorithms, K-Nearest Neighbor (KNN) and Logistic Regression (LR), for detecting fraudulent transactions on the Sparkov dataset (1.85 million transactions; a stratified subsample of 100,000 rows; 0.52% fraud rate), and analyses the effect of the Synthetic Minority Over-sampling Technique (SMOTE). Preprocessing includes temporal feature engineering, haversine distance, leak-free per-card behavioural features, one-hot and label encoding, and z-score standardisation. Models are evaluated on a stratified 80:20 split using the confusion matrix, accuracy, precision, recall, F1-score, ROC-AUC, and PR-AUC, complemented by decision-threshold tuning, confidence intervals over five repetitions, and the McNemar test. No single model dominates across all metrics. At the default threshold, KNN baseline achieves the highest F1 (0.407) and precision (0.540), while LR baseline achieves the highest PR-AUC (0.246); LR+SMOTE leads on recall (0.712) and ROC-AUC (0.861) but with very low precision (0.025). Threshold tuning lets LR baseline reach the best F1 (0.422 at a 0.071 cut-off). McNemar shows the KNN–LR difference is not significant at baseline (p = 0.282) but significant under SMOTE (p < 0.001). The main finding is that under severe imbalance ROC-AUC can be misleading and PR-AUC is more informative; KNN baseline is a balanced detector without tuning, threshold-tuned LR baseline gives the best single operating point, and LR+SMOTE suits cases where recall is the priority.
Felicia Sword, Christopher Andreas· UNP Journal of Statistics an...· 0 citations
Financial fraud detection is a critical challenge in modern banking systems, where fraudulent
transactions represent less than 0.13% of total transactions, creating severe class imbalance.
Traditional rule-based systems struggle to adapt to evolving fraud patterns, necessitating
machine learning approaches that can learn complex patterns from historical data while
handling extreme class imbalance. This study implements and evaluates three feature selection
techniques Minimum Redundancy-Maximum Relevance (mRMR), Multi Spatially Uniform
ReliefF (MultiSURF), and Hilbert-Schmidt Independence Criterion (HSIC) combined with eight
optimized machine learning classifiers including Support Vector Machine (SVM), Logistic
Regression (LR), K-Nearest Neighbors (KNN), Decision Tree (DT), AdaBoost (Ada), Bagging
(Bag), Stacking (Stack), and Voting classifiers. The financial payment services fraud dataset
containing 6,362,620 transactions was preprocessed using under sampling for class balancing
and Standard Scaler for normalization. GridSearchCV with 5-fold cross-validation was
employed for hyperparameter optimization across 32 experimental configurations. Experimental
results demonstrate that Bagging classifier without feature selection achieves the highest
performance with 99.51% accuracy, 99.15% precision, 99.88% recall, 99.51% F1-score, and
99.92% ROC-AUC in 10.67 seconds training time. Among feature selection methods, MultiSURF
maintains competitive performance, while mRMR and HSIC show performance degradation.
Ensemble methods consistently outperform single classifiers across all experimental scenarios.
This study provides empirical evidence that for fraud detection with well-engineered features in
low-dimensional spaces, ensemble methods yield optimal results. The undersampling strategy
proves effective for handling class imbalance while maintaining computational efficiency. The
findings demonstrating that feature selection is not always beneficial when original features are
already highly relevant.
Richard Chafukira Phiri· IIARD INTERNATIONAL JOURNAL...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
