Skip to content

Latest commit

 

History

History
28 lines (26 loc) · 2.34 KB

README.md

File metadata and controls

28 lines (26 loc) · 2.34 KB
Raw

Project Goals

  1. Showcase the computer science skills and knowledge that I have acquired with this degree.
  2. Design a project that solves a business problem, and define parameters/measures of success.
  3. Use machine learning to solve the business problem.
  4. Create meaningful data visualizations that enrich the presentation of this project.
  5. Design a user-friendly interface with interactive queries to use the product.

Project Constraints

  1. The machine learning model(s) must be properly designed, trained, and applied. The program must include a means to measure accuracy.
  2. The project must show the ability to support featurizing, parsing, cleaning, and wrangling datasets.
  3. Industry standard practices for data security must be followed.

Project Summary

  • I designed an ebook recommendation system for a fictitious company.
  • The business problem was to design software that could provide accurate and relevant ebook recommendations to the users of an online ebook marketplace.
  • The dataset was collected from Kaggle. It contains anonymized user demographic information, so no further action was required to make the dataset more secure.
  • Data ETL was performed. Pandas dataframes were used to load the data in from CSV files.
  • I employed various data-cleaning methods. Certain data entries and features were excluded based on the criteria that I set.
    • Entries that excluded key user demographic information were omitted from the cleaned dataset.
    • Features of the dataset that provided no value were omitted. Examples include book ISBN, URL links to pictures of the books, etc.
  • I used two machine learning models.
    • Matrix Factorization was used for the collaborative filtering of the cleaned data.
    • K-means clustering was used to cluster similar books into groups based on the latent relationships found with Matrix Factorization. Recommendations for a user were generated from these groups.
  • 3 key visualizations were used:
    • A bar graph showing the distribution of book ratings before and after the dataset is sampled.
    • A heat map to show the results of training the matrix factorization model.
    • A cluster graph to show the groups created by the K-means clustering model.
  • The GUI was created with a jupyter notebook widget. A user simply has to select a book, and then related suggestions are generated for other books to read.