hoda233 / financial-risk-modelling-of-european-p2p-investment-platform.

About The main purposes of this analysis are to summarize the characteristics of variables that can affect the loan status and to get some ideas about the relationships among variables.

Financial-risk-modelling-of-European-P2P-investment-platform

About

About The main purposes of this analysis are to summarize the characteristics of variables that can affect the loan status and to get some ideas about the relationships among variables.

Problem

Supervised machine learning problem.

• Binary classification:

  • Loan Status.

• Regression:

  • EMI (Equated Monthly Installments)
  • ELA (Eligible Loan Amount)
  • ROI (Return on Investment)

Libraries and packages

conda install numpy conda install pandas pip install seaborn

conda install matplotlib conda install scikit-learn

conda install -c conda-forge xgboost pip install streamlit

Structure

• Loading Packages and Data
• Data Exploration
• Data Preprocessing
  • Drop unneeded features
  • For Classification: Handle target feature (Status)
  • For Regression: Generate targets features (EMI, ELA, ROI)
  • Handle missing values
  • Handle outliers
• EDA (Exploratory Data Analysis)
  • Univariate Analysis
  • Bivariate Analysis & Correlation Analysis
• Encoding Categorical Variables
• Scaling
• Feature Selection
• PCA (Principal Component Analysis)
• Separate Modeling
  • For Classification:
    • Logistic Regression
    • Random Forest Classification
    • XGB Classification
    • Gaussian Naive Bayes
  • For Regression:
    • Linear Regression
    • Ridge Regression
    • Random Forest Regression

Final Pipelines

Each pipelines includes Scaling, PCA, Model.

pipeline = Pipeline([
    ('scaler', StandardScaler()),
    ('pca', PCA(n_components =  20)),
    ('classifier', model)
])

• Models used in Regression pipelines:

Model	R2 Scores for each target	R2 Score
LinearRegression	[0.76497051 0.57446454 0.39988935]	0.5797748017895326
Ridge	[0.7649705 0.57446456 0.39988934]	0.5797748025381682
RandomForestRegressor	[0.92705843 0.71706524 0.8533169 ]	0.8324801907433681
XGBRegressor	[0.93815903 0.71667844 0.85145203]	0.8354298342198373

• Models used in Classification pipelines:

Model	R2 Scores for each target
KNeighborsClassifier	0.9311306028395153
XGBClassifier	0.9449936817066825
LogisticRegression	0.9276741247305433
RandomForestClassifier	0.9417602021853861

Deployment

• Choose XGBRegressor, XGBClassifier as the final pipelines to be used in the app
• Create a streamlit app to access model and get predictions

How to run

• Dataset file should be in path dataset/Bondora_raw.csv
• Run Final_Pipeline.ipynb notebook to get the models
• To run the app: run python -m streamlit run app.py in vscode terminal

Application Screenshots

Numerical inputs

Categorical inputs

Outputs