This task will be building a machine learning model that attempts to predict whether a loan from LendingClub will become high risk or not.
LendingClub is a peer-to-peer lending services company that allows individual investors to partially fund personal loans, as well as buy and sell notes backing the loans on a secondary market. LendingClub offers their previous data through an API.
The data from LendingClub loans will be used to create machine learning models to classify the risk level of given loans. This task will compare the Logistic Regression model and Random Forest Classifier.
The loan datasets are located in the /Resources directory.
The code and analysis is located in the main directory and the filenane is Credit Risk Evaluator.ipynb
- The training dataset has been created from the 2019 loans using
pd.get_dummies()to convert the categorical data to numeric columns. - The same is done for 2020 loans, also using `pd.get_dummies().
- Missing categories were added into the testing set.
Rather than fitting a linear line as done in linear regression model, The logistic regression model creates an S-shaped curve fitting the data. This is a binary claffication algorithm that can be used to model the probability of a certain class or event.
Random forest is a classification algorithm that consists of a large number of individual decision trees that operate together. The individual trees in the random forest spits out a class prediction and the class with the most votes becomes the model’s prediction. It's use of bagging and feature randomness when building each individual tree creates an uncorrelated forest of trees whose prediction is more accurate than that of an individual tree.
The Preprocessing data has been revisted and StandardScaler has been applied to scale the training and testing sets.
The LogisticRegression and RandomForestClassifier models have been fitted and scored using the scaled data.
LendingClub (2019-2020) Loan Stats. Retrieved from: https://resources.lendingclub.com/