/AmazonBestSellerPrediction

Research for Data Science Project - Best Seller products on Amazon, how can they be predicted?

Primary LanguageJupyter Notebook

Amazon Best Seller Prediction

What makes an item in Amazon a Best Seller ?

In our project we try to find out why an item is a Best Seller Jupyter Notebook File .ipynb: https://github.com/yanivlav/AmazonBestSellerPrediction/blob/master/AmazonBestSellersPrediction.ipynb

🌟 Highlights

  • Data Acquisition
  • Data Handling
  • Exploratory Data Analysis ( EDA )
  • Supervised Learning
  • Unsupervised Learning

ℹ️ Overview

Data Acquisition

  • Using Selenium we managed to scrape more the 100,000 data cells on more than 13,000 items in Amazon Both Best Seller and regular items

Data Handling

  • With the BeautifulSoup ( bs4 ) library we collected and arranged the data and exported the data into a .csv file
  • We used various methodes in order to clean tha data :
    • Removing duplicates
    • Removing corrupted values
    • Normalizing
    • Transform all data to type Float

Exploratory Data Analysis ( EDA )

  • In oredr to analyze our DataFrame we used seaborn and matplotlib visualization libraries

    • Pie plot
    • Histogram
    • Boxplot
    • Scatter plot
    • pairplot

  • Dealing with outliers using IQR variable

  • Using Chi Test to determine if there is correlation between data features

Supervised Learning

  • prediction models :
    • Logistic Regression, r2_score
    • KNN

Unsupervised Learning

  • prediction models :
    • K-means, silhouette score
    • Hierarchical agglomerative clustering
    • DBSCAN

📑 Conclusion

After using Both Supervised and UnSupervised Learning we can conclude that we can't really find out what makes an item a BestSeller for sure. We tried to predict a bestseller using classification, we've got a right prediction but we can't really find it exact every time. Using KNN was not so helpful also, we tried to figure what k as number of classes we got best score for k=1. We also tried split the items into different groups using the unsupervised methods and didn't really find good results

More information is needed on items in order to get better results

✍️ Authors

  • Yaniv Lavi

✉️ yan465@gmail.com
💿 https://github.com/yanivlav
🖥️ https://www.linkedin.com/in/yanivlav1/

  • Stas Bratanitch

✉️ bstasb@gmail.com
💿 https://github.com/StasBratanich
🖥️ https://www.linkedin.com/in/stas-bratanich-computer-science/