Differentiated thyroid cancer (DTC), comprising papillary and follicular thyroid cancers, is the most prevalent type of thyroid malignancy. Accurate prediction of DTC is crucial for improving patient outcomes. Machine learning (ML) offers a promising approach to analyze risk factors and predict cancer recurrence. In this study, we aimed to develop predictive models to identify patients at an elevated risk of DTC recurrence based on 16 risk factors. We developed six ML models and applied them to a DTC dataset. We evaluated the ML models using Synthetic Minority Over-Sampling Technique (SMOTE) and with hyperparameter tuning. We measured the models’ performance using precision, recall, F1 score, and accuracy. Results showed that Random Forest consistently outperformed the other investigated models (KNN, SVM, Decision Tree, AdaBoost, and XGBoost) across all scenarios, demonstrating high accuracy and balanced precision and recall. The application of SMOTE improved model performance, and hyperparameter tuning enhanced overall model effectiveness.

With the growing popularity of cryptocurrencies, detecting potential market manipulation and fraudulent activities has become crucial for maintaining market integrity. In this study, we aim to detect anomalous Bitcoin transactions using an integrated approach by combining clustering techniques with statistical outlier detection. More specifically, anomalies were detected using three approaches: a distance-based method, flagging points with distances greater than the 95th percentile from their cluster centers; a statistical method, identifying transactions with any feature having an absolute Z-score greater than 3; and a hybrid approach, where transactions flagged by either method were considered anomalous. Using sample subset Bitcoin transaction data from 2015, our results showed that the combined approach was able to achieve the best performance with a total of 6492 (6.61%) detected anomalous transactions out of a total of 98,151 transactions.

Differentiated thyroid cancer (DTC), comprising of papillary and follicular thyroid cancers, is the most prevalent type of thyroid malignancy. Accurate prediction of DTC is crucial for improving patient outcomes. Machine learning (ML) offers a promising approach to analyze risk factors and predict cancer recurrence. In this study, we aimed to develop predictive models to identify patients at an elevated risk of DTC recurrence based on 16 risk factors. We developed six ML models and applied them to a DTC dataset. We evaluated the ML models using Synthetic Minority Over-Sampling Technique (SMOTE) and with hyperparameter tuning. We measured the models’ performance using precision, recall, F1 score, and accuracy. Results showed that Random Forest consistently outperformed the other investigated models (KNN, SVM, Decision Tree, AdaBoost, and XGBoost) across all scenarios, demonstrating high accuracy and balanced precision and recall. The application of SMOTE improved model performance and hyperparameter tuning enhanced overall model effectiveness.

Machine learning has become a central framework for analyzing high-dimensional biological data generated by genomics, transcriptomics, proteomics, single-cell sequencing, and multi-omics technologies. However, predictive modeling is frequently limited by severe class imbalance, where biologically important populations including rare disease subtypes, uncommon molecular states, and low-frequency cell populations are underrepresented. Synthetic oversampling addresses this challenge by increasing minority-class representation, with the Synthetic Minority Oversampling Technique (SMOTE) being the most widely used approach. However, conventional SMOTE assumes that geometric proximity reflects biological similarity, an assumption often violated in high-dimensional, nonlinear, and heterogeneous biological data. This review examines clustering-guided synthetic generation methods that integrate unsupervised learning before oversampling to preserve underlying biological structure. We evaluate partition-based, density-aware, fuzzy, cluster-filtering, and representation-based approaches, emphasizing their mathematical foundations, computational assumptions, and applications in molecular classification, rare population analysis, single-cell modeling, and multi-omics integration. Key challenges including cluster instability, preservation of rare biological populations, synthetic data validation, and distinguishing statistical rarity from biological significance are discussed. Finally, we outline emerging directions involving biological foundation models, graph-based learning, multi-omics representation spaces, and biologically constrained generative frameworks. These advances represent a shift from conventional data balancing toward structure-aware synthetic modeling that preserves meaningful biological organization while improving predictive performance.