Improving clustering-based and adaptive position-aware interpolation oversampling for imbalanced data classification DOI Creative Commons
Yujiang Wang, Marshima Mohd Rosli, Norzilah Musa

et al.

Journal of King Saud University - Computer and Information Sciences, Journal Year: 2024, Volume and Issue: 36(10), P. 102253 - 102253

Published: Dec. 1, 2024

Language: Английский

Impact of Nature of Medical Data on Machine and Deep Learning for Imbalanced Datasets: Clinical Validity of SMOTE Is Questionable DOI Creative Commons
Seifollah Gholampour

Machine Learning and Knowledge Extraction, Journal Year: 2024, Volume and Issue: 6(2), P. 827 - 841

Published: April 15, 2024

Dataset imbalances pose a significant challenge to predictive modeling in both medical and financial domains, where conventional strategies, including resampling algorithmic modifications, often fail adequately address minority class underrepresentation. This study theoretically practically investigates how the inherent nature of data affects classification classes. It employs ten machine deep learning classifiers, ranging from ensemble learners cost-sensitive algorithms, across comparably sized datasets. Despite these efforts, none classifiers achieved effective dataset, with sensitivity below 5.0% area under curve (AUC) 57.0%. In contrast, similar applied dataset demonstrated strong discriminative power, overall accuracy exceeding 95.0%, over 73.0%, AUC above 96.0%. disparity underscores unpredictable variability data, as exemplified by dispersed homogeneous distribution among other classes principal component analysis (PCA) graphs. The application synthetic oversampling technique (SMOTE) introduced 62 patients based on merely 20 original cases, casting doubt its clinical validity representation real-world patient variability. Furthermore, post-SMOTE feature importance analysis, utilizing SHapley Additive exPlanations (SHAP) tree-based methods, contradicted established cerebral stroke parameters, further questioning coherence augmentation. These findings call into question SMOTE underscore urgent need for advanced techniques innovations predicting minority-class outcomes datasets without depending strategies. approach developing methods that are not only robust but also clinically relevant applicable scenarios. Consequently, this future research efforts bridge gap between theoretical advancements practical, applications models like healthcare.

Language: Английский

Citations

9
Predicting learning achievement using ensemble learning with result explanation DOI Creative Commons
Tingting Tong, Zhen Li

PLoS ONE, Journal Year: 2025, Volume and Issue: 20(1), P. e0312124 - e0312124

Published: Jan. 2, 2025

Predicting learning achievement is a crucial strategy to address high dropout rates. However, existing prediction models often exhibit biases, limiting their accuracy. Moreover, the lack of interpretability in current machine methods restricts practical application education. To overcome these challenges, this research combines strengths various algorithms design robust model that performs well across multiple metrics, and uses analysis elucidate results. This study introduces predictive framework for based on ensemble techniques. Specifically, six distinct are utilized establish base learner, with logistic regression serving as meta learner construct an predicting achievement. The SHapley Additive exPlanation (SHAP) then employed explain Through experiments XuetangX dataset, effectiveness proposed verified. outperforms traditional deep terms results demonstrate learning-based significantly methods. feature importance analysis, SHAP method enhances improves reliability results, enabling more personalized interventions support students.

Language: Английский

Citations

0
A non-parameter oversampling approach for imbalanced data classification based on hybrid natural neighbors DOI

Junyue Lin,

Lu Liang

Applied Intelligence, Journal Year: 2025, Volume and Issue: 55(5)

Published: Jan. 22, 2025

Language: Английский

Citations

0
Do they like your game? Early-stage churn prediction using a two-phase neural network system DOI
Thu‐Huong Thi Hoang, Nguyen Tan Cam

Engineering Applications of Artificial Intelligence, Journal Year: 2025, Volume and Issue: 144, P. 110102 - 110102

Published: Jan. 25, 2025

Language: Английский

Citations

0
A novel instance density-based hybrid resampling for imbalanced classification problems DOI
You‐Jin Park, Chung-Kang Ma

Soft Computing, Journal Year: 2025, Volume and Issue: 29(4), P. 2031 - 2045

Published: Feb. 1, 2025

Language: Английский

Citations

0
MWMOTE-FRIS-INFFC: An Improved Majority Weighted Minority Oversampling Technique for Solving Noisy and Imbalanced Classification Datasets DOI Creative Commons
Dong Zhang, Xiang Huang, Gen Li

et al.

Applied Sciences, Journal Year: 2025, Volume and Issue: 15(9), P. 4670 - 4670

Published: April 23, 2025

In view of the data fault diagnosis and good product testing in industrial field, high-noise unbalanced samples exist widely, such are very difficult to analyze field analysis. The oversampling technique has proved be a simple solution past, but it no significant resistance noise. order solve binary classification problem data, an enhanced majority-weighted minority technique, MWMOTE-FRIS-INFFC, is introduced this study, which specially used for processing noise-unbalanced classified sets. method uses Euclidean distance assign sample weights, synthesizes combines new into with larger weights belonging few classes, thus solves scarcity smaller class clusters. Then, fuzzy rough instance selection (FRIS) eliminate subsets synthetic low clustering membership, effectively reduces overfitting tendency caused by oversampling. addition, integration fusion iterative filters (INFFC) helps mitigate noise issues, both raw On basis, series experiments designed improve performance 6 algorithms on 8 sets using MWMOTE-FRIS-INFFC algorithm proposed paper.

Language: Английский

Citations

0
Geometric Relative Margin Machine for Heterogeneous Distribution and Imbalanced Classification DOI
Xiaojing Lv, Ling-Wei Huang, Yuan‐Hai Shao

et al.

Published: Jan. 1, 2024

Class imbalance and heterogeneous data distribution pose significant challenges in classification tasks across various real-world applications. Addressing these issues, this paper introduces the Geometric Relative Margin Machine (GRMM), a novel model that innovatively merges strategies of with advanced adjustment techniques. GRMM is specifically designed to effectively manage dual class heterogeneity. Empirical evaluations on benchmark datasets practical scenarios reveal not only significantly improves accuracy but also enhances robustness against diverse distributions. This study underscores efficacy navigating complexities varied sizes distributions, showcasing its potential as superior tool for complex problems.

Language: Английский

Citations

0
An Improved Hybrid Sampling Method for Classifying imbalanced Data to Predict Student Performance DOI
Mohamed Bellaj, Ahmed Bendahmane,

Said Boudra

et al.

Published: May 2, 2024

Language: Английский

Citations

0
Overlap to equilibrium: Oversampling imbalanced datasets using overlapping degree DOI
Sidra Jubair, Jie Yang, Bilal Ali

et al.

Information Processing & Management, Journal Year: 2024, Volume and Issue: 62(2), P. 103975 - 103975

Published: Nov. 23, 2024

Language: Английский

Citations

0
Improving clustering-based and adaptive position-aware interpolation oversampling for imbalanced data classification DOI Creative Commons
Yujiang Wang, Marshima Mohd Rosli, Norzilah Musa

et al.

Journal of King Saud University - Computer and Information Sciences, Journal Year: 2024, Volume and Issue: 36(10), P. 102253 - 102253

Published: Dec. 1, 2024

Language: Английский

Citations

0