project3_wine

Project Description

• As a data science consultant Team BCR will find the drivers of wine quality for the California Wine Institute. We will find clusters within the data for exploration, understanding and modeling to present to the data science team for winery supply chain marketing. The project will culminate in a presentation to the data science team.

Project Goals

• Discover features/clusters helpful in predicting wine quality

• Use features to develop a machine learning model to predict the quality of wine on a scale of 1-10

• Other key drivers:

  • Is alcohol associated with quality?
  • Is chlorides associated with quality?
  • Is residual_sugar associated with quality?
  • Is alcohol associated with density?

Initial Thoughts

• There are some key indicators in the data that may predict the 'quality' of wine and that those indicators will be evident by the conclusion of the project.

The Plan

• Acquire: build organization repository named Team BCR and invite members. Pull winequality-red.csv and winequality-white.csv from data.world into repository. Build .gitignore and env.py files.

• Prepare the data using the following columns:

  • target: 12 - quality (score between 0 and 10)
  • features:
    • 1 - fixed acidity
    • 2 - volatile acidity
    • 3 - citric acid
    • 4 - residual sugar
    • 5 - chlorides
    • 6 - free sulfur dioxide
    • 7 - total sulfur dioxide
    • 8 - density
    • 9 - pH
    • 10 - sulphates
    • 11 - alcohol

• Explore dataset for predictors of wine 'quality'

  • Answer the following questions:
  • Is alcohol associated with quality?
  • Is chlorides associated with quality?
  • Is residual_sugar associated with quality?
  • Is alcohol associated with density?

• Develop a model

  • Using the selected data features develop appropriate predictive models
  • Evaluate the models in action using train and validate splits as well as scaled data
  • Choose the most accurate model
  • Consider several clusters to enhance the chosen model
  • Evaluate the most accurate model using the final test data set
  • Draw conclusions

Data Dictionary

Feature	Definition (measurement)
Fixed Acidity	The fixed amount of tartaric acid. (g/L)
Volatile Acidity	A wine's acetic acid; (High Volatility = High Vinegar-like smell). (g/L)
Citric Acid	The amount of citric acid; (Raises acidity, Lowers shelf-life). (g/L)
Residual Sugar	Leftover sugars after fermentation. (g/L)
Chlorides	Increases sodium levels; (Affects color, clarity, flavor, aroma). (g/L)
Free Sulfur Dioxide	Related to pH. Determines how much SO2 is available. (Increases shelf-life, decreases palatability). (mg/L)
Total Sulfur Dioxide	Summation of free and bound SO2. (Limited to 350ppm: 0-150, low-processed, 150+ highly processed). (mg/L)
Density	Between 1.08 and 1.09. (Insight into fermentation process of yeast growth). (g/L)
pH	2.5: more acidic - 4.5: less acidic (range)
Sulphates	Added to stop fermentation (Preservative) (g/L)
Alcohol	Related to Residual Sugars. By-product of fermentation process (vol%)
Quality	Score assigned between 0 and 10; 0=low, 10=best

Steps to Reproduce

1. Clone the repo git@github.com:Team-BCR/project3_wine.git in terminal
2. Use personal env.py to connect
3. Run notebook
4. Use final_report for findings

Takeaways and Conclusions

Models used:

• Regression:
  • Ordinary Least Squares (OLS)
  • LassoLars
  • Polynomial Regression
  • Generalized Linear Model (GLM)

Exploration Conclusions: * The mean wine quality score is 5.87 * all bottles had a score from 1 to 10 * No feature correlated more than a .5 correlation coefficient * alcohol was the highest correlation with a .46

Modeling Conclustions:

• Interestingly, sending in all, features even those not correlated with the target, led to the best performing model
• Clustering did not help.

Recommendations

• Create wines that optimize the chemical properties as described above
  • i.e. the higher/lower values that correspond with higher quality scores
  • This will not guarantee a higher quality wine, but it should increase the probability of creating a higher quality wine

Citation

P. Cortez, A. Cerdeira, F. Almeida, T. Matos and J. Reis. Modeling wine preferences by data mining from physicochemical properties. In Decision Support Systems, Elsevier, 47(4):547-553. ISSN: 0167-9236.