Repository for the data science wine analysis and rating prediction
I was looking for educational wine dataset with understandable features and suitable for creating ML model for my first DS project. I couldn't find anything relevant, so decided to scrap data from Vivino.com
Data contains 4 files for each winestyle: red, white, rose and sparkling. Also there is a file with wine varieties for further analysis.
Analyzing data presented on Vivino.com, I noticed that there are no bottles that have less than 25 ratings, apparently because the company considers the rating of such wines is not accurate enough. From the graphic below you can see main problem of the project. Number of Ratings has the exponential distribution and there a lot of wines, which has low Number of Ratings. There is a real problem for business, because the main reason of Vivino app is to give info about quality of wine to customers, but for huge number of wines there are no rating at all. My idea is to apply ML methods to predict rating of wines wich has less than 25 ratings.
As it turned out, the problem of reviews distribution is exists in many spheres. It consists in the fact that customers are often afraid of choosing a product or service that no one has ever bought before. Due to this, many businesses lose large amounts of money on the unnormal distribution of customers by product. My idea is to create a model for any such business that predicts the rating based on other features, it can help to increase the demand for new, but promising products
As a result, I found out consumer behavior strategy and made a model which shows cross validation accuracy = 0.13 in rating prediction.