Hossein Azizi

Hossein Azizi

Certified Data Scientist, Ph.D. Computational Support Developer at Natural Resources Canada

Forecasting Online Sales

less than 1 minute read

Summary

E-com Fitness is a web-based application designed to forecast fitness product sales using data extracted from social media and time-series analysis of historical data.

Data extraction:

  • Scraping data from social media (e.g., posts, comments, reactions…etc.)
  • Scraping product history
  • Utilize new web scraping tools (Octoparse), crawling from different APIs
  • Scraped 10000 Facebook posts/tweets; form 100 top fitness pages; 10000 sold products from last three months on Ebay.ca; historical data from last three year for 1000 unique products on Walmart, Canadian Tire, and Best buy Handling blocking and scrape from ajax websites

Feature Engineering:

  • create new features based on inventory history, detecting & removing outliers based on percentile, bucketizing price & dates

Construct a predictive model:

  • Train and benchmark a Decision-Tree based algorithm

Delivering Model:

  • Deploying on Heroku using Flask. Click to visit the deployed app.

You may also enjoy

Image Classification

2 minute read

Classifying images using series of neural networks.

Social Graph

1 minute read

Build a social graph of social connections of the population, and used that to determine influential people within the group.