Online shoppers purchasing intention prediction

An analysis of a shoppers purchasing intention dataset. Contains a Python Notebook with visualizations, data pre-processing and machine/deep learning models for supervised learning. A model is then exported to a Django application to be used with a REST API. Requests can be made with cURL or every request senders to make predictions on a new observation on his intention to purchase or not.

Context

ESILV Project - Python for data analysis course 2020

TU Klarissa, THIOL Antoine

Github

Dataset description

The dataset used is the "Online Shoppers Purchasing Intention Dataset".

It's a dataset regrouping online sessions and for each one, whether or not it ended with shopping. Of the 12,330 sessions in the dataset, 84.5% (10,422) were negative class samples that did not end with shopping, and the rest (1908) were positive class samples ending with shopping.

The dataset consists of 10 numerical and 8 categorical attributes. The 'Revenue' attribute is a boolean feature indicating if the session ended with shopping or not.

Dataset Link

Citations: Sakar, C.O., Polat, S.O., Katircioglu, M. et al. Neural Comput & Applic (2018)

Data analysis

Study objectives

Our objective is, for a given observation, to be able to predict whether or not the session will end with shopping. As the target is a boolean variable, it's a binary classification problem. Using the multiple features of a session, we will implement data-driven models to predict the shopper intention.

Data-driven models

For this dataset, we used and studied multiple machine/deep learning models. We showed an evolution of our models, the advantages and disadvantages and we compared them. The three models used are:

• a Decision Tree
• a Multi-layer Perceptron
• a Random Forest Classifier

Django application and REST API

The Django application allows to use the exported model with a REST API. We can store session information, display, modify or delete them. We can also predict the "Revenue" class of our dataset which will tell us if, given the session information, the user will purchase or not.

Technologies

• Python 3.8
• Django, Jupyter Notebook
• pandas, numpy, scikit-learn, keras, matplotlib, seaborn, joblib

User installation

First, we have to start the Django Web application.

❯ cd /path/to/project/apirest
❯ python manage.py runserver

REST endpoints

GET /shoppers/

Fetch the database population.

❯ curl localhost:8000/shoppers/

POST /shoppers/

Save a new shopper in the database.

> curl -X POST -H "Content-Type: application/json" localhost:8000/shoppers/ -d 
"{\"Administrative\":0,
  \"Administrative_Duration\":0.0,
  \"Informational\":0,
  \"Informational_Duration\":0.0,
  \"ProductRelated\":1,
  \"ProductRelated_Duration\":0.0,
  \"BounceRates\":0.2,
  \"ExitRates\":0.2,
  \"PageValues\":0.0,
  \"SpecialDay\":0.0,
  \"Month\":\"Feb\",
  \"OperatingSystems\":1,
  \"Browser\":1,
  \"Region\":1,
  \"TrafficType\":1,
  \"VisitorType\":\"Returning_Visitor\",
  \"Weekend\":false,
  \"Revenue\":null}"

GET /shopper/:id

Fetch a specific shopper.

❯ curl -H "Accept: application/json" localhost:8000/shopper/1/
{
  "Administrative":0,
  "Administrative_Duration":0.0,
  "Informational":0,
  "Informational_Duration":0.0,
  "ProductRelated":1,
  "ProductRelated_Duration":0.0,
  "BounceRates":0.2,
  "ExitRates":0.2,
  "PageValues":0.0,
  "SpecialDay":0.0,
  "Month":"Feb",
  "OperatingSystems":1,
  "Browser":1,
  "Region":1,
  "TrafficType":1,
  "VisitorType":"Returning_Visitor",
  "Weekend":false,
  "Revenue":null
}

PUT /shopper/:id

Modify a specific shopper.

> curl -X PUT -H "Content-Type: application/json" localhost:8000/shopper/4/ -d 
"{\"Administrative\":123456,
  \"Administrative_Duration\":0.0,
  \"Informational\":0,
  \"Informational_Duration\":0.0,
  \"ProductRelated\":1,
  \"ProductRelated_Duration\":0.0,
  \"BounceRates\":0.2,
  \"ExitRates\":0.2,
  \"PageValues\":0.0,
  \"SpecialDay\":0.0,
  \"Month\":\"Feb\",
  \"OperatingSystems\":1,
  \"Browser\":1,
  \"Region\":1,
  \"TrafficType\":1,
  \"VisitorType\":\"Returning_Visitor\",
  \"Weekend\":false,
  \"Revenue\":null}" 

DELETE /shopper/:id

Delete a specific shopper.

> curl -X DELETE localhost:8000/shopper/4/

POST /predict/

Predict the purchasing intention for a given session.

> curl -X POST -H "Content-Type: application/json" localhost:8000/predict/ -d
"{\"Administrative\":0,
  \"Administrative_Duration\":0.0,
  \"Informational\":0,
  \"Informational_Duration\":0.0,
  \"ProductRelated\":1,
  \"ProductRelated_Duration\":0.0,
  \"BounceRates\":0.2,
  \"ExitRates\":0.2,
  \"PageValues\":0.0,
  \"SpecialDay\":0.0,
  \"Month\":\"Feb\",
  \"OperatingSystems\":1,
  \"Browser\":1,
  \"Region\":1,
  \"TrafficType\":1,
  \"VisitorType\":\"Returning_Visitor\",
  \"Weekend\":false,
  \"Revenue\":null}"
	 
{
  "Administrative": 123456, 
  "Administrative_Duration": 0.0, 
  "Informational": 0, 
  "Informational_Duration": 0.0, 
  "ProductRelated": 1, 
  "ProductRelated_Duration": 0.0, 
  "BounceRates": 0.2, 
  "ExitRates": 0.2, 
  "PageValues": 0.0, 
  "SpecialDay": 0.0, 
  "Month": "Feb", 
  "OperatingSystems": 1, 
  "Browser": 1, 
  "Region": 1, 
  "TrafficType": 1, 
  "VisitorType": "Returning_Visitor", 
  "Weekend": false, 
  "Revenue": false
}

License

MIT