Comprehensive Analysis of Random Forest and XGBoost Performance with SMOTE, ADASYN, and GNUS Under Varying Imbalance Levels

Technologies, Journal Year: 2025, Volume and Issue: 13(3), P. 88 - 88

Published: Feb. 20, 2025

This study examines the efficacy of Random Forest and XGBoost classifiers in conjunction with three upsampling techniques—SMOTE, ADASYN, Gaussian noise (GNUS)—across datasets varying class imbalance levels, ranging from moderate to extreme (15% 1% churn rate). Employing metrics such as F1 score, ROC AUC, PR Matthews Correlation Coefficient (MCC), Cohen’s Kappa, this research provides a comprehensive evaluation classifier performance under different scenarios, focusing on applications telecommunications domain. The findings highlight that tuned paired SMOTE (Tuned_XGB_SMOTE) consistently achieves highest score robust across all levels. emerged most effective method, particularly when used XGBoost, whereas performed poorly severe imbalance. ADASYN showed effectiveness but underperformed Forest, GNUS produced inconsistent results. underscores impact data imbalance, MCC, scores fluctuating significantly, AUC remained relatively stable. Moreover, rigorous statistical analyses employing Friedman test Nemenyi post hoc comparisons confirmed observed improvements PR-AUC, MCC were statistically significant (p < 0.05), Tuned_XGB_SMOTE significantly outperforming Tuned_RF_GNUS. While differences ROC-AUC not significant, consistency these results multiple reliability our framework, offering validated attractive solution for model selection imbalanced classification scenarios.

Language: Английский

---

Ensemble-based customer churn prediction in banking: a voting classifier approach for improved client retention using demographic and behavioral data

Discover Sustainability, Journal Year: 2025, Volume and Issue: 6(1)

Published: Jan. 14, 2025

Language: Английский

---

Decoding PM2.5 Prediction in Nanning Urban Area, China: Unraveling Model Superiorities and Drawbacks Through SARIMA, Prophet, and LightGBM

Algorithms, Journal Year: 2025, Volume and Issue: 18(3), P. 167 - 167

Published: March 14, 2025

With the rapid development of industrialization and urbanization, air pollution is becoming increasingly serious. Accurate prediction PM2.5 concentration great significance to environmental protection public health. Our study takes Nanning urban area, which has unique geographical, climatic source characteristics, as object. Based on dual-time resolution raster data China High-resolution High-quality Dataset (CHAP) from 2012 2023, carried out using SARIMA, Prophet LightGBM models. The systematically compares performance each model spatial temporal dimensions indicators such mean square error (MSE), absolute (MAE) coefficient determination (R2). results show that a strong ability mine complex nonlinear relationships, but its stability poor. obvious advantages in dealing with seasonality trend time series, it lacks adaptability changes. SARIMA based series theory performs well some scenarios, limitations non-stationary heterogeneity. research provides multi-dimensional reference for subsequent predictions, helps researchers select models reasonably according different scenarios needs, new ideas analyzing change patterns, promotes related field science.

Language: Английский

---

Realistic Data Delays and Alternative Inactivity Definitions in Telecom Churn: Investigating Concept Drift Using a Sliding-Window Approach

Applied Sciences, Journal Year: 2025, Volume and Issue: 15(3), P. 1599 - 1599

Published: Feb. 5, 2025

Predicting customer churn is essential for telecommunications companies to maintain profitability. However, training models on historical leads performance degradation when they are applied future conditions—a phenomenon known as concept drift. We employ a sliding-window approach that separates the and testing time windows, creating future-based “true test”. Using unique real data, we show CatBoost classifier model trained older data can remain relevant new, unseen intervals used. A key innovation of our work use 40-day “partial churn” labels; these labels accurately predicts 90-day by simply adjusting decision threshold. Out six modeled scenarios, in main realistic scenario, retained an accuracy above 0.798 F1 near 0.704, reflecting its robustness even under real-world delays potential Overall, findings emphasize do not necessarily “expire” with time; rather, their varies according tested. This research underscores importance truly evaluation (instead artificial splits) offers practical guidance earlier detection facing delays.

Language: Английский

---

Enhancing customer retention with machine learning: A comparative analysis of ensemble models for accurate churn prediction

International Journal of Information Management Data Insights, Journal Year: 2025, Volume and Issue: 5(1), P. 100331 - 100331

Published: Feb. 27, 2025

Language: Английский

---

Bridging Predictive Insights and Retention Strategies: The Role of Account Balance in Banking Churn Prediction

AI, Journal Year: 2025, Volume and Issue: 6(4), P. 73 - 73

Published: April 10, 2025

The banking industry faces significant challenges, from high customer churn rates to threatening long-term revenue generation. Traditionally, models assess service quality using satisfaction metrics; however, these subjective variables often yield low predictive accuracy. This study examines the relationship between attrition and account balance decision trees (DT), random forests (RF), gradient-boosting machines (GBM). research utilises a dataset applies synthetic oversampling class distribution during preprocessing of financial variables. Account is primary factor in predicting churn, as it yields more accurate predictions compared traditional assessment methods. tested model set achieved its highest performance by applying boosting evaluation data highlights critical role indicators shaping effective retention strategies. By leveraging machine learning intelligence, banks can make informed decisions, attract new clients, mitigate risk, ultimately enhancing results.

Language: Английский

---

Mitigating class imbalance in churn prediction with ensemble methods and SMOTE

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: May 9, 2025

This study examines how imbalanced datasets affect the accuracy of machine learning models, especially in predictive analytics applications such as churn prediction. When are skewed towards majority class, it can lead to biased model performance, reducing overall effectiveness. To analyze this impact, research utilizes a dataset evaluate data imbalance influences accuracy. The utilized nine individual classifiers along with six homogeneous ensemble models effects on performance. Single classifier struggle identify underlying patterns data, while ensembles improve performance by focusing minority class. However, when trained unbalanced their remains subpar. top were selected for further investigation based data. A SMOTE sampling technique was employed create balanced dataset, ensuring that all classes adequately represented. generated model's improved from 61 79%, indicating removal bias target results showed Adaboost, an optimal classifier, demonstrated superior F1-Score 87.6% identifying potential and assessing customer account health. findings emphasize importance accurate ML predictions.

Language: Английский

---

Multi-Layer Perceptron and Radial Basis Function Networks in Predictive Modeling of Churn for Mobile Telecommunications Based on Usage Patterns

Applied Sciences, Journal Year: 2024, Volume and Issue: 14(20), P. 9226 - 9226

Published: Oct. 11, 2024

Customer retention is a key priority for mobile telecommunications companies, as acquiring new customers significantly more costly than retaining existing ones. A major challenge in this field predicting customer churn—users discontinuing services. Traditional predictive models such rule-based systems often struggle with the complex, non-linear nature of behavior. To address this, we propose use deep learning techniques, specifically multi-layer perceptron (MLP) and radial basis function (RBF) networks, to improve accuracy churn predictions. However, while neural networks excel performance, they are criticized being “black-box” models, lacking interpretability. real-world data set considered, which originally contained information about 15,000 randomly selected clients. Various network structures configurations analyzed. The obtained results compared generated using fuzzy rough-set systems. MLP model achieved an almost perfect 0.999 F-measure 0.989, outperforming traditional methods Although RBF slightly lagged accuracy, it demonstrated superior recall 0.993, indicating better identification potential churners. These demonstrate that enhance performance modeling. interpretability also discussed since bears significance real applications. Our contribution lies showing prediction though remains critical area future work.

Language: Английский

---

Strategic management of employee churn: Leveraging machine learning for sustainable development and competitive advantage in emerging markets

Business Strategy & Development, Journal Year: 2024, Volume and Issue: 7(4)

Published: Nov. 5, 2024

Abstract Employee churn or attrition presents significant challenges, especially in emerging markets, where it can disrupt business operations and inflate recruitment costs. This research leverages machine learning techniques to predict employee churn, focusing on developing sustainable inclusive retention strategies that enhance competitiveness. By analyzing a range of predictive algorithms key variables associated with the study identifies most effective models for predicting attrition. A comprehensive exploratory data analysis was conducted using an indigenous model, offering practical insights human resource management markets. The findings align development goals (SDGs), promoting decent work, economic growth. contributes strategy by proposing data‐driven solutions workforce stability development.

Language: Английский

---

Integrating Genetic Algorithms and Analytic Hierarchy Process for Customer Retention Optimization in E-Commerce

Published: Sept. 7, 2024

Language: Английский

---