Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

Transforming financial documents into credit decisions using explainable artificial intelligence and optical

Sachin Malave1, Bharti Khemani2, Hrishit Patil3

  • 1Head of Computer Engineering Department, A. P. SHAH Institute of Technology, Survey No 12, 13, Opp. Hypercity Mall, Kasarvadavali, Ghodbunder Road, Thane West, Thane, Maharashtra 400615, India.

Methodsx
|June 3, 2026
PubMed
Summary

Related Concept Videos

Non-equilibrium in the Cell01:16

Non-equilibrium in the Cell

An important concept in studying metabolism and energy is that of chemical equilibrium. Most chemical reactions are reversible. They can proceed in both directions, releasing energy into their environment in one direction, and absorbing it from the environment in the other direction. The same is true for the chemical reactions involved in cell metabolism, such as the breaking down and building up of proteins into and from individual amino acids, respectively. Reactants within a closed system...
Mathematical Modeling: Problem Solving01:29

Mathematical Modeling: Problem Solving

Mathematical modeling transforms real-world scenarios into mathematical expressions, allowing for structured problem-solving and analysis. This process involves defining the situation, assigning variables to measurable quantities, selecting an appropriate model, and solving the resulting equation. Such models are invaluable in finance, providing precise methods to evaluate investments, loans, and repayment structures.A widely used example is the calculation of fixed monthly payments on a loan,...

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Improved graph convolutional network for emotion analysis in social media text.

MethodsX·2025
Same author

Detecting health misinformation: A comparative analysis of machine learning and graph convolutional networks in classification tasks.

MethodsX·2024
See all related articles

This study introduces a new automated system that reads financial documents using optical character recognition and then uses advanced machine learning to make credit decisions. By applying specific interpretability tools, the system explains why a credit decision was made, which helps banks meet regulations and build trust with borrowers. The results show that a specific model called XGBoost is the most accurate at predicting credit risk compared to several other common methods.

Area of Science:

  • Financial technology research within Explainable Artificial Intelligence systems
  • Computational linguistics and document processing in banking analytics

Background:

No prior work has fully resolved the tension between high-predictive accuracy and model transparency in automated lending. Prior research has shown that complex machine learning algorithms often function as opaque black boxes. This uncertainty drove concerns regarding regulatory compliance and borrower fairness in modern banking. It was already known that traditional statistical methods lack the performance of ensemble techniques. That gap motivated the development of systems that prioritize both precision and interpretability. Scholars have previously identified that digital transformation in finance necessitates robust data extraction from unstructured documents. This study addresses the challenge of integrating document parsing with advanced predictive modeling. The current landscape requires a balance between computational power and human-understandable decision logic.

Purpose Of The Study:

The aim of this study is to develop an integrated framework for automated credit scoring that utilizes document-based data extraction. The researchers seek to resolve the conflict between high-predictive accuracy and model transparency in lending. They address the problem where complex algorithms often operate as opaque systems, which hinders regulatory compliance. The authors investigate how optical character recognition can transform unstructured financial documents into usable structured features. They intend to demonstrate that interpretability tools can provide clear justifications for automated credit decisions. The team explores the performance of various machine learning models to identify the most effective approach for risk prediction. They strive to enhance borrower trust by providing both local and global explanations for lending outcomes. This work motivates a shift toward more accountable and reliable automated financial services.

Keywords:
Credit scoringExplainable artificial intelligence (XAI)Financial document processingLIMEMachine learningOptical character recognition (OCR)SHAPXGBoostmachine learning financeautomated lending systemspredictive modeling bankingdocument data extraction

Frequently Asked Questions

The researchers propose an integrated framework using Optical Character Recognition to extract data, followed by XGBoost for prediction and SHAP/LIME for interpretability. This combination allows the system to achieve 92.5% accuracy while providing clear reasons for each credit decision.

The authors utilize EasyOCR, PyPDF2, and PdfPlumber to parse applicant files. These tools convert unstructured document formats into structured data, which then undergoes normalization and feature engineering before entering the machine learning pipeline.

The researchers indicate that interpretability tools are necessary to address concerns regarding fairness and regulatory compliance. Without these explanations, complex ensemble models function as opaque systems, making it difficult for financial institutions to justify their lending choices to borrowers or auditors.

Related Experiment Videos

Main Methods:

The review approach utilizes a multi-stage pipeline for processing applicant records. Investigators employ specialized libraries to parse portable document formats and image-based files. They perform rigorous data cleaning and normalization to ensure input quality. The team executes structured feature engineering to prepare variables for supervised learning. They implement various ensemble and deep learning algorithms to evaluate predictive capability. Researchers apply specific interpretability modules to generate both global and local decision insights. They calculate multiple performance metrics to validate the robustness of the proposed system. This design ensures that the final output remains both accurate and interpretable for end users.

Main Results:

Key findings from the literature indicate that the XGBoost model achieves the highest predictive accuracy at 92.5%. The analysis shows that LightGBM follows with 90.8% accuracy in credit risk tasks. Logistic Regression demonstrates 91% accuracy, while Random Forest reaches 87%. The Long Short-Term Memory network shows the lowest performance among the tested models at 81%. Researchers report that the integration of document parsing significantly improves the reliability of the input data. The performance metrics confirm that the proposed framework maintains high precision and recall across all tested scenarios. These results highlight the effectiveness of combining structured features with advanced machine learning for automated lending.

Conclusions:

The authors propose that their integrated framework successfully bridges the gap between predictive performance and model transparency. They suggest that combining structured financial features with explainable techniques improves overall decision-making reliability. The team claims that their approach supports necessary regulatory compliance standards within the banking sector. They observe that the XGBoost model demonstrates superior predictive capabilities compared to alternative algorithms tested. The researchers indicate that local and global explanations provide meaningful insights into credit scoring outcomes. They maintain that this methodology enhances trust by clarifying the logic behind automated lending choices. The findings imply that document-based data extraction is a viable pathway for modernizing credit risk assessment. The study concludes that transparency is a manageable requirement when utilizing advanced machine learning in finance.

The team employs structured financial features derived from applicant documents. This data type serves as the input for supervised learning models, ensuring that the predictive engine relies on verified information extracted directly from the provided identity and financial records.

The study measures performance using accuracy, precision, recall, F1-score, Mean Absolute Error, and Root Mean Square Error. These metrics confirm that XGBoost outperforms LightGBM, Random Forest, Logistic Regression, and Long Short-Term Memory networks in credit risk prediction.

The authors claim that their approach improves decision-making reliability. They propose that by making automated processes transparent, financial institutions can better align with legal standards while simultaneously increasing the accuracy of their risk assessments.