Metric Marvins stands out as the premier credit analyst, leveraging the prowess of historical data and advanced machine learning techniques. Our state-of-the-art model excels in accurately predicting creditworthiness, adapting seamlessly to economic fluctuations, and outperforming traditional scoring methods.
Credit score cards are widely used in the financial industry to evaluate the risk of issuing credit cards. Traditional methods can be affected by economic fluctuations, prompting the use of machine learning algorithms. While these improve accuracy, they often lack transparency in explaining acceptance or rejection decisions to customers and regulators. This Project aims to improve transparency in the whole process quantifying the reasons why one is approved or denied a credit card
Existing credit scoring methods based on historical data and traditional financial metrics face limitations in predicting creditworthiness accurately due to economic fluctuations. There is a need to develop a credit scoring model that leverages historical data and advanced machine learning techniques for better prediction while providing transparent results for customers and regulators. The goal is to create a reliable and transparent credit scoring system that informs decision-making and ensures fair treatment of applicants.
The end user is the bank that will use this deployment in conjunction with their own to complement the credit scoring system
- Develop a credit scoring model that incorporates personal and social factors and machine learning algorithms to enhance the accuracy of creditworthiness predictions.
- Improve transparency in credit scoring by utilizing interpretable machine learning techniques, allowing for clear explanations of acceptance or rejection decisions to customers and regulatory bodies.
- Mitigate the impact of economic fluctuations on credit scoring models by incorporating dynamic factors and adapting the model to changing economic conditions.
- Optimize the balance between prediction accuracy and interpretability to ensure a fair and reliable credit scoring system for both applicants and financial institutions.
We set a specific target for our credit card approval prediction model, aiming to achieve a recall score of 80%. This means our focus is on maximizing the identification of approved credit card applications, minimizing the chances of false negatives and ensuring a high level of accuracy in predicting positive outcomes.
For this project, we utilized a dataset sourced from Kaggle.There are two datasets,the Applications Dataset and the Credits Dataset. The Applications dataset has 438557 rows and 18 columns. The Credits dataset has 1048575 rows and 3 columns. The features encompass social and economic factors that have an impact on the credit card application status of individual applicants
The data for this project was sourced from Kaggle . This data consists of information of past application and credits. For more info [Data Description](data\Data Description.txt)
During our thorough analysis of the raw data, we discovered missing values specifically in the occupation variable. To address this, we performed imputation techniques to fill in the missing values. Additionally, we conducted outlier detection to identify and handle any extreme values present in the dataset.
Furthermore, we addressed duplicated entries by removing them to ensure data integrity. To enhance clarity and consistency, we renamed variables appropriately. In order to establish uniformity, we converted certain variables to a yearly basis. Additionally, we adjusted the data types of variables as needed to ensure accurate and efficient analysis.
We conducted an in-depth analysis of the dataset to uncover patterns, trends, and insights. This included analyzing the Status of the clients based their previous loan payments.
C - indicates that the client has paid off the loan in full for that particular month. X- suggests that the client did not have any loan obligations for that specific month Overdue ,write offs- indicating bad debts or write-offs that have been pending for more than 150 days. The others indicates the client's payment is overdue by certain number of days
We used five algorithms for our modeling: Decision Tree Classifier(baseline model), Random Forest Classifier, Ada Boost Classifier, Gradient Boost Classifier and XGB Classifier
We opted for the Decision Tree Classifier as our final model. This choice was driven by its interpretability, which enables us to gain insights into the decision-making process and provide transparent explanations for credit card application outcomes.
The decision tree classifier attained a recall score 0.85, test accuracy of 0.79 and train accuracy of 0.87.
While XGB Class delivered an impressive recall score, ensuring a higher number of true positive predictions, our incorporation of Decision Tree has further improved the overall prediction accuracy.
Our Minimum Viable Product for credit card approval prediction combines the power of Decision Tree with the interpretability of SHAP (SHAPLEY ADDITIVE EXPLANATIONS). Decision Tree provides accurate predictions, while SHAP helps us understand the influence of each feature in the approval decision. This integration ensures not only precise credit card approval predictions but also transparency and fairness in the decision-making process.
-
After the models were evaluated, the Decision Tree model was selected as the final model based on high performance on evaluation metrics. Training Accuracy: 0.870 Testing Accuracy: 0.789 Recall Score: 0.85 were achieved by the final model, indicating its ability to classify approval or denial status with high accuracy and low false positives and false negatives.
-
The dataset was highly imbalanced, with a small percentage of denial status, with the majority being approved on their credit card application.
-
It was observed that the most common status is "C" (likely_good_debtor), followed by "X" (likely_good_debtor) and "0" (Neutral_Debtor). Features such as income type, property ownership, gender, and family status showed relationships with the credit approval status.
-
Based on the bank's risk appetite, debtors can be categorized as "Approved" (likely_good_debtor, Neutral_Debtor) or "Denied" (likely_Bad_Debtor, likely_written_off) based on the credit approval status.
-
Further explore the reasons behind different loan statuses ("C," "X," "0," etc.) to gain insights into loan approval decisions.
-
Conduct additional research or collect more data to better understand the demographic and social economic factors contributing to loan approval.
-
Explore the use of additional data sources, such as social media activity or online transaction history, to improve credit assessment and enhance predictive accuracy.
-
Establish a process for ongoing evaluation and refinement of the credit scoring model, including regular updates and retraining based on new data and feedback.
-
Conduct training and educational programs to enhance stakeholders' understanding of the credit scoring system, promoting transparency and empowering customers to make informed financial decisions.
- The dataset used has limited number of features while there could be other factors that contribute to the credit approval status.
- The quality and availability of data can significantly impact the project. Limitations may include incomplete or missing data, data with errors or inconsistencies, biased or unrepresentative data, or limited access to relevant data sources.
- The evolving nature of data science introduces the risk of project outcomes becoming outdated as new methods, techniques, and data sources emerge.
- Different algorithms have inherent limitations, such as assumptions about data distribution, sensitivity to outliers, or constraints on features or samples, which can impact the accuracy and effectiveness of the results.
- New features can be generated from the existing data to provide more insights.
- Explore alternative data sources or consider data augmentation techniques to enhance the quality and availability of the data.
- Consider incorporating more complex modeling techniques or exploring ensemble methods to capture nuances and avoid oversimplification.
- Regularly review and keep up-to-date with advancements in algorithms and techniques to leverage new approaches that overcome limitations.
- Regularly reassess and update models or approaches to incorporate new methods, algorithms, or data sources as they become available.
├── data <- Folder containing datasets used for the project
|
├── .gitignore <- Ignore unnecessary files and directories
|
├── LICENSE.md <- The license for the project
|
├── README.md <- The top-level README for reviewers of this project
|
├── approval.ipynb <- Narrative documentation of analysis in Jupyter Notebook
|
├── presentation.pdf <- PDF version of project presentation
|
└── write_up.pdf <- PDF version of project write up
To install and run this project, follow these steps:
Certainly! Here are the installation and running steps in bullet point format:
Clone the repository to your local machine using the following command:
git clone https://github.com/ndungek/Credit-Card-Approval-Prediction
Navigate to the project's root directory:cd Credit-Card-Approval-Prediction
Install the project dependencies:pip install jupyter
Open your web browser and access the Jupyter Notebook interface at http://localhost:8888.
From the Jupyter Notebook interface, navigate to the notebook file (.ipynb) and open it.
On your terminal run streamlit run app.py
Feel free to reach out to the collaborators mentioned in the project if you have any questions or need further assistance.
Leah Katiwa: @katiwaleah26@gmail.com
Maureen Kitang'a: @ndungek66@gmail.com
Ronald Nyagaka: nyagaka@outlook.com
Paul Kamau: @kamaa.njuguna16@yahoo.com
Beatrice Kirui: @beatriceckirui15@gmail.com
Cliff Shitote: @cliffbrecht6@gmail.com