Project Goal

Find drivers of wine quality and create a model that can accurately predict wine quality while performing better than baseline.

Project Description

Using the dataset for wine quality from Data World, look for any effects of physicochemical properties of wine on its resulting quality.

Initial Questions/Thoughts

We believe that density, pH, alcohol, and sulphates will be the best predictors of wine quality.

Data Dictionary

Column Name	Description	Key
fixed_acidity	the difference between total acidity and volatile acidity	milliequivalents per liter (int)
volatile_acidity	measure of volatile acids (acids that readily evaporate)	grams of acetic acid per liter (int)
citric_acid	weak organic acid naturally occurring in fruits
residual_sugar	amount of sugar remaining after fermentation	grams per liter (int)
chlorides	amount of salt in wine
free_sulfur_dioxide	free form of SO$\2$, used as preservative
total_sulfur_dioxide	total of free SO$\2$ and SO$\2$ bound to other chemicals
density	mass per unit volume of wine at 20°C
pH	measure of wine from basic to acidic	[0-14], wines are usually between [3-4]
sulphates	additive that contributes to SO$\2$ levels
alcohol	percent alcohol content of wine	percentage (float)
quality	quality of wine (numeric)	[0-9]
wine_type	type of wine	[Red, White]
quality_bin	quality of wine (category)	[Good, Bad]

Steps to Reproduce

Data was acquired from: https://data.world/food/wine-quality. Clone the repo and run through the final report.

Project Plan

Acquire

• Acquire dataset from link
• Add additional column for wine_type
• Merge two dfs into one to create full df
• Cache full df for future use
• View data .info, .describe, .shape

Prepare

• Adjust column names to be Python-readable
• View/correct datatypes
• Handle nulls (potential imputing)
• Visualize full dataset for univariate exploration (histograms and boxplots)
  • Handle outliers
• Get rid of unnecessary columns
• Splitting data

Pre-processing

• Scaling data on train
• Encoding any necessary columns
• Document how we're changing the data

Exploration

• Use unscaled data for multivariate exploration
  • Hypothesize
  • Visualize
  • Run stats tests
    • Run chi-squared test on clusters vs. target
  • Summarize
• Potential Feature Engineering

Exploration Summary

Initial Questions/Thoughts

Modeling

• Use scaled/encoded data
• Split into X_variables and y_variables
• Determine evaluation metrics
  • Establishing baseline
• Run different models on train/validate
• Pick best model and evaluate on test

Conclusion

Recommendations