Leakage-controlled Crowdfunding Success Prediction: Do Feature Selection and SMOTE Improve Boosting-Based Models?
Abstract
Decision-making in digital financial information systems plays an important role in predicting crowdfunding success for investors, platform providers, and campaign creators. Ignoring data imbalance and the potential for information leaks in the pursuit of high accuracy actually backfires. This kind of approach creates the illusion of performance that looks perfect on paper but is actually unreliable in practice. This study proposes a leakage-controlled framework that combines Hybrid Feature Selection, RFECV, SMOTE, and boosting algorithms using a public Turkish crowdfunding dataset. The dataset contains 76.9% unsuccessful and 23.1% successful campaigns. Pearson Correlation and Mutual Information were used for feature selection, followed by RFECV for feature optimization. The data splitting stage was intentionally completed before feature filtering, and class balancing via SMOTE was applied only to the training data set to ensure that the evaluation remained objective without data leakage. Separate data testing showed that the XGB configuration, supported by GBM-RFECV and SMOTE, achieved the best results Precision, Recall, F1-score, ROC-AUC, and PR_AUC of 0.871, 0.906, 0.888, 0.975, 0,923, along side minority successful campaign. The results suggest that leakage-controlled feature selection and imbalance handling can provide a more realistic assessment of model performance for live crowdfunding campaign monitoring.
Related
- Gradient boosting machine for survival prediction in studies with general failure-time outcome-dependent sampling design.
- Reproducibility and Leakage-Controlled Evaluation of Machine-Learning Models for Chronic Kidney Disease Classification
- Leakage-Controlled Projected Quantum Kernels for Orbital Conjunction Risk Assessment Using SOCRATES Data
- Model selection in linear mixed-effects analysis of craving outcomes in a randomized controlled trial of medications for opioid use disorders
- FDR-Controlled Variable Selection for Generalized Linear Models and Cox Regression with Virtual Dummies
BibTeX
@article{azzahra2026leakagecontrolled,
title = {{Leakage-controlled Crowdfunding Success Prediction: Do Feature Selection and SMOTE Improve Boosting-Based Models?}},
author = {Adellia Nirma Azzahra and Riska Dhenabayu and Dhiyan Septa Wihara and Muhammad Jiwa Akmal Sani Arman Rachman Liga},
journal = {Journal of Information Systems and Informatics},
year = {2026},
month = oct,
doi = {10.63158/journalisi.v8i5.1760},
url = {https://doi.org/10.63158/journalisi.v8i5.1760},
note = {CC BY},
}
RIS
TY - JOUR
TI - Leakage-controlled Crowdfunding Success Prediction: Do Feature Selection and SMOTE Improve Boosting-Based Models?
AU - Adellia Nirma Azzahra
AU - Riska Dhenabayu
AU - Dhiyan Septa Wihara
AU - Muhammad Jiwa Akmal Sani Arman Rachman Liga
PY - 2026
DA - 2026/10/24
JO - Journal of Information Systems and Informatics
DO - 10.63158/journalisi.v8i5.1760
UR - https://doi.org/10.63158/journalisi.v8i5.1760
AB - Decision-making in digital financial information systems plays an important role in predicting crowdfunding success for investors, platform providers, and campaign creators. Ignoring data imbalance and the potential for information leaks in the pursuit of high accuracy actually backfires. This kind of approach creates the illusion of performance that looks perfect on paper but is actually unreliable in practice. This study proposes a leakage-controlled framework that combines Hybrid Feature Selection, RFECV, SMOTE, and boosting algorithms using a public Turkish crowdfunding dataset. The dataset contains 76.9% unsuccessful and 23.1% successful campaigns. Pearson Correlation and Mutual Information were used for feature selection, followed by RFECV for feature optimization. The data splitting stage was intentionally completed before feature filtering, and class balancing via SMOTE was applied only to the training data set to ensure that the evaluation remained objective without data leakage. Separate data testing showed that the XGB configuration, supported by GBM-RFECV and SMOTE, achieved the best results Precision, Recall, F1-score, ROC-AUC, and PR_AUC of 0.871, 0.906, 0.888, 0.975, 0,923, along side minority successful campaign. The results suggest that leakage-controlled feature selection and imbalance handling can provide a more realistic assessment of model performance for live crowdfunding campaign monitoring.
SN - 2656-5935
SN - 2656-4882
ER -