Insurance cross-selling potential prediction:

Working on a dataset with imbalanced target classes

Libraries

• Tabular

  • pandas
  • numpy

• Tools

  • joblib: persistence

• Visualization

  • plotly express / graphing objects
  • ipywidgets: subplot container

• Machine Learning

  • scikit-learn: pipeline, models
  • imbalanced-learn: resampling
  • feature-engine: transformers

• API

  • flask

Structure

.
├── app : API 
│   └── templates: html structure of the form page
├── bin: dataset downloader (requires Kaggle API key)
├── data: train / test + answer
├── models: ml models and pipelines
├── notebooks: exploration, evaluation
├── output: plots
├── src: pipeline builder, viz functions, etc
└── tests: api, result matches local / distant models

API

Run locally

export FLASK_APP=app/app.py

flask run -h 0.0.0.0 -p 5000

API call example

curl -X POST \
  -d "Gender=male" \
  -d "Vehicle_Damage=yes" \
  -d "Vehicle_Age=<1 Year" \
  -d "Region_Code=28" \
  -d "Policy_Sales_Channel=124" \
  -d "Annual_Premium=30000" \
  -d "Age=25" \
  -d "Previously_Insured=0" \
  http://localhost:5000/predict

UI app

http://127.0.0.1:5000/predict

Viz