[Paper Review] Analyzing Machine Learning Models for Credit Scoring with Explainable AI and Optimizing Investment Decisions

Swati Tyagi|arXiv (Cornell University)|Sep 19, 2022

Financial Distress and Bankruptcy Prediction24 citations

TL;DR

The paper compares multiple ML models for credit scoring, applying LIME and SHAP for explainability, and investigates ML-driven investment strategies to maximize profitability with reduced risk.

ABSTRACT

This paper examines two different yet related questions related to explainable AI (XAI) practices. Machine learning (ML) is increasingly important in financial services, such as pre-approval, credit underwriting, investments, and various front-end and back-end activities. Machine Learning can automatically detect non-linearities and interactions in training data, facilitating faster and more accurate credit decisions. However, machine learning models are opaque and hard to explain, which are critical elements needed for establishing a reliable technology. The study compares various machine learning models, including single classifiers (logistic regression, decision trees, LDA, QDA), heterogeneous ensembles (AdaBoost, Random Forest), and sequential neural networks. The results indicate that ensemble classifiers and neural networks outperform. In addition, two advanced post-hoc model agnostic explainability techniques - LIME and SHAP are utilized to assess ML-based credit scoring models using the open-access datasets offered by US-based P2P Lending Platform, Lending Club. For this study, we are also using machine learning algorithms to develop new investment models and explore portfolio strategies that can maximize profitability while minimizing risk.

Motivation & Objective

• Motivate the use of explainable AI in financial ML applications such as credit underwriting and investment decisions.
• Evaluate a range of ML models from simple classifiers to ensembles and neural networks for credit scoring.
• Assess post-hoc explainability methods (LIME, SHAP) on credit scoring models.
• Develop ML-based investment models and portfolio strategies to maximize profitability while minimizing risk.
• Highlight the applicability of open-access Lending Club data for model evaluation.

Proposed method

• Compare single classifiers (logistic regression, decision trees, LDA, QDA), heterogeneous ensembles (AdaBoost, Random Forest), and sequential neural networks on credit scoring tasks.
• Apply post-hoc model-agnostic explainability techniques (LIME and SHAP) to assess model explanations.
• Utilize Lending Club open-access datasets to evaluate credit scoring models.
• Develop and test ML-based investment models and portfolio strategies aimed at profitability and risk minimization.
• Report qualitative and qualitative observations on model performance and explainability.

Experimental results

Research questions

• RQ1 Which ML models provide the best performance for credit scoring on the Lending Club data?
• RQ2 How do ensemble methods and neural networks compare to single classifiers in this setting?
• RQ3 How effective are LIME and SHAP in explaining credit scoring models?
• RQ4 Can ML-based investment models improve profitability while reducing risk compared to baseline strategies?

Key findings

• Ensemble classifiers and neural networks outperform single classifiers in credit scoring.
• LIME and SHAP are applied to evaluate explainability of the credit scoring models.
• The study uses Lending Club open-access datasets for evaluation.
• The work also develops ML-based investment models and portfolio strategies to optimize profitability with risk considerations.