Enhancing Classification Performance in Imbalanced Datasets Using GANified-SMOTE with Latent Factor
摘要
Imbalanced class distribution is one major issue in building robust predictive models. This is because minority samples are difficult to identify by conventional methods, which are biased towards the majority class. Such a problem is exceptionally vital in high-stakes applications like fraud detection and healthcare, where misclassifying rare occurrences would have significant implications. This study addresses the issue of class imbalance in predictive modeling by integrating GANified-SMOTE and latent factors into three classification algorithms: Random Forest, Gradient Boosting and Decision Tree. The new approach generates high-quality synthetic samples to improve the minority classes’ representation in five benchmark datasets. According to experimental results, Random Forest always outperforms other classifiers. Using the credit card fraud detection dataset, it has an accuracy of 99.99%. Its rival, Gradient Boosting, however, with 98.79% accuracy in minority classes identification, indicates that it lags behind. Decision Tree classifier at 99.90% accuracy, on the other hand, uses a balanced performance. All these observations indicate that the suggested GANified-SMOTE approach greatly promotes classifier performance and introduces a good option for producing useful predictive models in imbalanced data environments. This has significant implications in high-stakes applications such as finance, fraud prevention and healthcare analytics.