This repository details an initial approach to developing a predictive model for the Lending Club loan data, available here.
In this project, we develop a model to detect if a loan will default before the loan is issued. Headline figures for the tested models on the full dataset:
| Metric | Logistic Regression | Decision Tree | Random Forest | Gaussian |
|---|---|---|---|---|
| AUC | 0.72 | 0.52 | 0.59 | 0.64 |
| Accuracy | 0.41 | 0.50 | 0.61 | 0.71 |
| Precision | 0.94 | 0.80 | 0.82 | 0.82 |
| Recall | 0.27 | 0.48 | 0.63 | 0.82 |
- Clone the repository:
$ git clone https://github.com/andmikey/lendingclub-analysis.git- Download the data and place in correct locations:
This is best done by hand, unless you already have the API tool set up.
Go to the link above, download the zip file to lending-club-loan-data.zip, unzip it, and move the files to their appropriate directories:
$ mkdir -p data/{raw,interim,processed}
$ mv loan.csv data/raw/
$ mv LCDataDictionary.xlsx references/
- Start a new pip environment and install the required packages:
$ virtualenv -p `which python3` env
$ source env/bin/activate
$ pip3 install -r requirements.txt- Pre-process the dataset for training:
$ make data N_SAMPLES=0- Train the model:
$ make train_model DATA=data/processed/loan.csv MODEL=models/model.pickle- (Optional) If there is a new unlabelled dataset, say at data/processed/new_data.csv, predict the labels for the new observations:
$ make predict_model MODEL=models/model.pickle INPUT=data/processed/new_data.csv OUTPUT=models/test_predict.csv- (Optional) You can also do piece-wise data creation: sampling the data (to reduce training size), cleaning the data, and adding features:
$ make sample_data N_SAMPLES=50000
$ make clean_data SRC=data/interim/loan_sampled_50000.csv DEST=data/interim/loan_sampled_50000-cleaned.csv
$ make add_features SRC=data/interim/loan_sampled_50000-cleaned.csv DEST=data/processed/loan_sampled_50000.csv├── Makefile <- Makefile with commands like `make data` or `make train`
├── README.md <- The top-level README for developers using this project.
├── data
│ ├── interim <- Intermediate data that has been transformed.
│ ├── processed <- The final, canonical data sets for modeling.
│ └── raw <- The original, immutable data dump.
│
├── models <- Trained and serialized models, model predictions, or model summaries
│
├── notebooks <- Jupyter notebooks. Naming convention is a date, username, and description.
│ └── exploratory <- Notebooks for exploring and testing approaches (cleaning, visualisation, prediction)
│
├── references <- Data dictionaries, manuals, and all other explanatory materials.
│
├── reports <- Generated analysis
│ └── figures <- Generated graphics and figures to be used in reporting
│
├── requirements.txt <- The requirements file for reproducing the analysis environment
│
├── setup.py <- makes project pip installable (pip install -e .) so src can be imported
├── src <- Source code for use in this project.
│ ├── __init__.py <- Makes src a Python module
│ │
│ ├── data <- Scripts to generate data
│ │ ├── clean_dataset.py
│ │ ├── make_dataset.py
│ │ └── sample_dataset.py
│ │
│ ├── features <- Scripts to turn raw data into features for modeling
│ │ └── build_features.py
│ │
│ ├── models <- Scripts to train models and then use trained models to make
│ │ │ predictions
│ │ ├── predict_model.py
│ │ └── train_model.py
│ │
│ └── visualization <- Scripts to create exploratory and results oriented visualizations
│ └── visualize.py
Project based on the cookiecutter data science project template. #cookiecutterdatascience