International Journal of Innovative Research in Computer Science and Technology
Year: 2026, Volume: 14, Issue: 6
First page : ( 30) Last page : ( 37)
Online ISSN : 2347-5552
DOI: 10.55524/ijircst.2026.14.6.4 |
DOI URL: https://doi.org/10.55524/ijircst.2026.14.6.4
This is an open-access article distributed under the terms of the Creative Commons Attribution 4.0 International License (CC BY 4.0).
http://creativecommons.org/licenses/by/4.0
Article Tools: Print the Abstract | Indexing metadata | How to cite item | Email this article | Post a Comment
Shivangi Srivastava , Raj Bihari Lal Srivastava
Microenterprises often have difficulty accessing formal finance due to limited credit history, incomplete financial records and lack of documented borrowing information. So, their financial situation and repayment ability may not be truly reflected by traditional credit evaluation methods. Opportunities in machine learning and artificial intelligence to improve credit risk assessment involve incorporating alternative financial and behavioral indicators. However, predictive models can make unreliable decisions when the information they have is incomplete, inconsistent, or insufficient. In this context, this paper proposes an uncertainty-aware AI framework for credit risk assessment of thin-file microenterprises with alternative financial behavioral indicators. The proposed framework consists of financial preprocessing of data, behavioral feature extraction, machine learning based risk prediction, uncertainty quantification, and human-in-the-loop decision support. The candidate prediction models considered are Logistic Regression, Random Forest and XGBoost. To benchmark the context, we consider a public dataset of credit-risk of SMEs from Bangladesh with 1,200 loan records and 27 structured features. The repository benchmark results using five-fold stratified cross-validation show that Logistic Regression has the highest ROC-AUC (0.900 ± 0.024) and PR-AUC (0.776 ± 0.045), while XGBoost has the lowest Brier score (0.174). Random Forest shows the relatively lower discrimination performance and the largest reported train-test gap. These results can serve as a comparative reference for candidate prediction model selection. The proposed mechanisms of probability calibration, conformal prediction and human-review have not yet been evaluated experimentally. The benchmark dataset is a representation of the general SMEs’ population, which does not prove that all are thin file microenterprises. Thus, the reported results do not serve as direct evidence of the target population’s performance. We put forth a framework which is a structure for developing more open, which includes and presents the uncertainty in results and responsible AI based credit assessment systems.
Assistant Professor, Department of Computer Application, C.M.P Degree College, Prayagraj, Uttar Pradesh, India
No. of Downloads: 3 | No. of Views: 66
Sarla, Yojna Arora.
Nov-Dec 2026 - Vol 14, Issue 6
Chandra Sekhar Sanaboina, Tammisetti Surekha.
Sep-Oct 2026 - Vol 14, Issue 5
Gunjan Mishra, Harsh Dev, Sudheer Kumar Singh.
Sep-Oct 2026 - Vol 14, Issue 5
