Skip to content

Latest commit

 

History

History
92 lines (70 loc) · 2.66 KB

README.md

File metadata and controls

92 lines (70 loc) · 2.66 KB

Website Hits Prediction

This is a project for predicting the hits on a web page in an IPython Notebook which demonstrates

  • Understanding Data
  • Exploratory Data Analysis
    • Visualization
    • Manipulation
    • Feature Engineering
  • Data Preparation
  • Model Selection
  • Validation

The goal of this repository is to demonstrate an insightful understanding of data using visualizations and feature engineering for the prediction model.

Quick Start: View a static version of the notebook in the comfort of your own browser.

Dependencies:

  • Python 2.7
  • Pandas
  • Sklearn
  • NumPy
  • Seaborn
  • Matplotlib
  • XGBoost

Installation

To run this notebook interactively:

  1. Clone this repo

    $ git clone https://github.com/techedlaksh/website-hits-prediction
$ cd website-hits-prediction

  2. Create new virtual environment

    $ sudo pip install virtualenv
$ virtualenv venv
$ source venv/bin/activate
$ pip install -r requirements.txt
 ```

  3. Run Notebook

    $ jupyter notebook

  4. Click on final-notebook.ipynb in the browser and enjoy!

  5. When you're done with notebook, close the jupyter from terminal and deactivate the virtual environment with deactivate.

Data

  • row_num: a nuber uniquely identifying each row.
  • locale: the platform of the session.
  • day_of_week: Mon-Fri, the day of the week of the session.
  • hour_of_day: 00-23, the hour of the day of the session.
  • agent_id: the device used for the session.
  • entry_page: describes the landing page of the session.
  • path_id_set: shows all the locations that were visited during the session.
  • traffic_type: indicates the channel the user cane through.
  • session_duration: the duration in seconds of the session.
  • hits: the number of interactions with the trivago page during the session.

Task

Use the data provided to build a model that predicts the number of hits per session, depending on the given parameters.

Evaluation

Predictions will be evaluated by the root mean square error.

Notebook covers these topics

Data Handling

  • Importing data with pandas
  • Understanding data using statistics with pandas
  • Exploring Data through Visualizations with Matplotlib
  • Feature Engineering
  • Data Preparation for the model

Model Selection

  • Logistic Regression
  • Random Forest
  • XGBoost
  • LightBGM

Valuation

  • K-folds cross validation to valuate results locally

Note: Trained models are exported which can be re-used by importing it into your script and predicting your data with the saved model.