A platform implemented using Machine learning models to make a real-time prediction using a particular given output.
LendingClub is a US based company and it is the world's biggest peer-to-peer lending platform. In this project, we aim to build a machine learning model that predicts the probability that a loan will be charged off or not. Such models help investors of LendingClub make better-informed decisions in terms of investment. We have used a 1.8 GB dataset with around 150 variables for each and every loan.
In training the model, we use only the features that are known to investors before they choose to invest in the loan. These key features include, the income of the borrower, FICO score, ratio of debt to income, and the required loan amount, the purpose,his location, grade, and the interest rate.
The modeling process involves several steps, consisting: removing the features with significant missing data, or that aren't exposed to investors; exploring the data, transforming the data, and visualizing it; creating dummy variables for categorical features; filling missing data with mean or median values; and the fitting models: logistic regression, random forest, gradient boosting. We have used cross validation techinique to select features which really influence a decision on the dataset and finalized 30 attributes for model training and testing.
All the analysis will be done in a Python Jupyter Notebook, utilizing the packages numpy, pandas, matplotlib, seaborn, and scikit-learn.
As a part of this project, We also developed an UI, which is an enterprise application aimed at the investors. An investor can enter the attributes in an UI and the response returned is a probability that the loan will be charged off.
IBM Watson, JavaScript, Python, React, Redux, REST API, Flask, Bootstrap, scikit-learn