Organization - The Mifos Initiative
Project Title - Scorecard for Credit Risk Assessment
Mentor - Lalit Mohan
- The project consisted of providing an AI powered solution to the users for credit assessment of loans. The project covered various aspects from classical AI, considering various statistical models, to the modern day neural network. The project is enriched with various credit modeling techniques, giving access to the user to choose one or any from them. It also takes care of the different data sources from which data can be fetched and has been fully incorporated to handle data coming from various sources like JSON/XML or SQL.
- During this period, I worked on configuring different sections, namely, rule based, statistical based and machine learning based models for credit risk assessment. Along with working on the backend, I worked in improving the already implemented to screens to cope up with the backend and provide additional functionalities to cover the need of the project.
- Initially, only the front-end was ready along with a basic implementation of the backend to make it functional. I implemented a rule based system that allows users to select features, according to their need, assigning them weights, and criteria based on which they would be evaluated. These rules are then taken into consideration while assessing credit risk. The final score is provided as an output with a color describing the intensity of risk, where green showing the least risk, amber showing mediocre, and red showing most. Not only a final color, but also colors with each of the features, with which rules were framed, are being provided to allow user to understand the reason behind the output.
- Incorporating different statistical methods like MANOVA, Value at Return, Regression.
- Completely Offline.
- Providing Accuracy as well as prediction of model
- Option to choose between different methods provided
- Providing all sorts of machine learning models, covering probalilstic, ensemble, neural network, etc.
- Running on all models and providing the output with best accuracy
- No need to re-train models again and again
- Saving the model showing best accuracy, so no need to run all models for output
- No need to store the dataset
- Dataset is loaded at runtime
- Just require a link where dataset is stored and data is fetched when required
- No need to upload the link every time to get predictions
- Added a data source option, with a text field to add the link from where data is required to be fetched. The link is then loaded at run time using the id against which the loan is being sanctioned.
- After fetching the data, it is then marked against the criterias created by the user to generate a final score in the rule based methodology.
- Initially the editing of features was not working, I implemented the backend code to allow the editing of features and configurations. This allows user to change the configurations and criterias according to the need, without changing the name of the feature.
- As the project deals with a lot of requests being made in and out of server, proper handling of errors is required, which was not done initially.
- Added filters and improved overall UI.
- Added processing of different types of inputs differntly, example, binary, interval based, etc.
- Corrected save feature implementation
- Corrected fetching of data from mifos database
- Added check for completion of feature form
- Use API's such as H2O AI for credit risk assessment
- Allow for fetching gaurded data
- Unit Testing
- Find the most appropriate models
- Implementing according to design pattern
- Compilation of scores from differnt sources in rule based methodology.
It was really amazing working with The Mifos Organization. It was so much to learn from them as a community as well as technology. I would really like to thank Ed Cable and my mentor Lalit Mohan for choosing me to work with them and also guiding me throughout the period. I learnt a lot from them. Also I would like to thanks the whole Mifos community as they were always there for help and guidance. It was a really great experience for me and a very informative too, apart from learning about the technology, I also learnt how to build production ready code as it is very important for one to know that and it would surely help me in my future.