Pei-Chun Lin

Machine learning (ML) model performance is often assessed by accuracy, but the quality and balance of data also play crucial roles. Imbalanced datasets, where the minority class has fewer samples than the majority class, can lead to biased predictions favoring the majority class. This study addresses the issue of class imbalance through resampling techniques, including random undersampling (RUS) and random oversampling (ROS), specifically applied to a fraud detection dataset. We classify the resampled datasets using random forest (RF) and gradient boosting (GB) models. Our findings indicate that the RF model, when combined with ROS, achieves an accuracy of 97.4%, surpassing the 96.1% accuracy of the GB model with RUS. This approach demonstrates the importance of addressing class imbalance to improve prediction accuracy in ML.

This study provides a comprehensive review on monocular metrology and fuzzy random-set portfolio management of financial assets. The findings and conclusions are elaborated as follows. Soft computing and AI have already enhanced and will further empower a variety of applications of monocular metrology and fuzzy random-set portfolio management of financial assets through progressive quantification and capturing of domain situations. The single most significant limitation of monocular metrology lies in its intrinsic incapability of direct measurement of 3D geometry through 2D imagery. The future of monocular metrology lies in deep learning and end-to-end solutions, multi-sensor data fusion, algorithmic optimization and real-time performance, self-supervised learning and generalization, and standardization and practical deployment. Neither statistical validation nor performance optimization alone is sufficient to support decision-makers in making portfolio decisions that are reliably and trust-worthy. A promising portfolio management decision-making framework should integrate the statistical rigor of fuzzy statistics with fuzzy random portfolio optimization techniques to quantitatively account for fuzziness and uncertainty while better balancing computational efficiency, statistical reliability, interpretability, and practical credibility.

One such serious problem in machine learning (ML) is imbalanced datasets. Minority class samples are usually sparse but hold significant meaning. The model can become biased toward the majority class due to unbalanced class distribution. This results in fraudulently high accuracy without being able to detect minority cases. This bias is also most perilous in critical applications, where ignoring minority cases can be highly destructive. To overcome this problem, the Synthetic Minority Oversampling Technique (SMOTE) is one of the most widely used. SMOTE creates balanced class distribution by interpolating between existing minority samples. It creates samples that are too close to one another and can lead to overfitting and limit the generalization of the model. Recent advancements in generative modeling, especially Generative Adversarial Networks (GANs), offer a more effective solution to handle class imbalance. GANs utilizes a generative discriminator structure to produce synthetic data highly similar to real data. A hybrid technique called GANified-SMOTE combines the power of SMOTE with the generation power of GANs to produce more diverse and realistic minority class samples. The technique improves the model strength and eliminates the limitations of traditional oversampling. This paper presents the incorporation of latent factors into the architecture of GANified-SMOTE framework. Latent variables reveal hidden structures and relations in the data, leading to a closer synthetic sample and improving classification accuracy. By incorporating latent factors, this research aims to build a better oversampling method for imbalanced classification sets.

Background: Imbalanced and overlapped data in customer churn prediction significantly impact classification results. Various sampling and hybrid sampling methods have demonstrated effectiveness in addressing these issues. However, these methods have not performed well with classical machine learning algorithms. Methods: To optimize the performance of classical machine learning on customer churn prediction tasks, this study introduces an extension framework called CostLearnGAN, a tabular generative adversarial network (GAN)-hybrid sampling method, and cost-sensitive Learning. Utilizing a cost-sensitive learning perspective, this research aims to enhance the performance of several classical machine learning algorithms in customer churn prediction tasks. Based on the experimental results classical machine learning algorithms exhibit shorter execution times, making them suitable for predicting churn in large customer bases. Results: This study conducted an experiment with six comparative sampling methods, six datasets, and three machine learning algorithms. The results show that CostLearnGAN achieved a satisfying result across all evaluation metrics with a 1.44 average mean rank score. Additionally, this study provided a robustness measurement for algorithms, demonstrating that CostLearnGAN outperforms other sampling methods in improving the performance of classical machine learning models with a 5.68 robustness value on average.

Multiclass classification in machine learning often faces significant challenges due to unbalanced datasets. This situation leads to biased predictions and reduced model performance. This research addresses this issue by proposing a novel approach that combines convolutional neural networks (CNNs) with class weights and early-stopping techniques. The motivation behind this study stems from the need to improve model performance, especially for minority classes, which are often neglected in existing methodologies. Although various strategies such as resampling, ensemble methods, and data augmentation have been explored, they frequently have limitations based on the characteristics of the data and the specific model type. Our approach focuses on optimizing the loss function via class weights to give greater importance to minority classes. Therefore, it reduces bias and improves overall accuracy. Furthermore, we implement early stopping to avoid overfitting and improve generalization by continuously monitoring the validation performance during training. This study contributes to the body of knowledge by demonstrating the effectiveness of this combined technique in improving multiclass classification in unbalanced scenarios. The proposed model is tested for oil palm leaves analysis to identify deficiencies in nitrogen (N), boron (B), magnesium (Mg), and potassium (K). The CNN model with three layers and a SoftMax activation function was trained for 200 epochs each. The analysis compared three scenarios: training with the imbalanced dataset, training with class weights, and training with class weights and early stopping. The results showed that applying class weights significantly improved the classification accuracy, with a trade-off in other class predictions. This indicates that, while class weight has a positive overall impact, further strategies are necessary to improve model performance across all categories in this study.

Credit score models are essential tools for evaluating creditworthiness and mitigating financial risks. However, the imbalanced nature of multi-class credit score datasets poses significant challenges for traditional classification algorithms, leading to poor performance in minority classes. This study explores the effectiveness of Generative Adversarial Network (GAN)-based oversampling methods, including CTGAN, CopulaGAN, WGAN-GP, and DraGAN, in addressing this issue. By synthesizing realistic data for minority classes and integrating it with majority class data, the study benchmarks these GAN-based methods across classical (KNN, Decision Tree, Logistic Regression) and ensemble machine learning models (XGBoost, Random Forest, LightGBM). Evaluation metrics such as accuracy and F1-score reveal that WGAN-GP consistently achieves superior performance, especially when combined with Random Forest, outperforming other methods in balancing dataset representation and enhancing classification accuracy. The results showed that WGAN-GP + RF achieved 0.873 in accuracy, 0.936 F1-score in the “good” class, 0.806 F1-score in the “poor” class, and 0.816 F1-score in the “standard” class. The findings underscore the potential of GAN-based oversampling in improving multi-class credit score classification and highlight future directions, including hybrid sampling and cost-sensitive learning, to address remaining challenges.