Prediction of Wine Quality based on Physicochemical Tests

• author: Jiajie (Joshua) Lim, Ling (Elina) Lin, Jiacheng Wang, Kangyu (Mark) Wang

This project aims to build machine learning models that predict a given wine’s quality based on its physicochemical characters. It uses classification and regression techniques k-nearest neighbors (k-NN) and support vector machines (SVMs) with RBF kernel.

Data

The data sets used in this project are of the prediction of wine quality based on physicochemical tests, related to red and white vinho verde wine samples, from the north of Portugal. The data were created by Paulo Cortez et al at the University of Minho, Guimarães, Portugal (2009). It was sourced from the UCI Machine Learning Repository (Dua and Graff 2017) and can be found here, specifically this file. The data provided only have physicochemical (inputs) and sensory (the output) variables available (e.g. there is no data about grape types, wine brand, wine selling price, etc.). There are a total of 4898 instances in our combined dataset of red wine and white wine.

Research Question

From the data set provided, we are interested in predicting the wine quality (white wine and red wine) based on the physicochemical tests. We are engaged in predicting the quality score (score between 0 and 10) of a wine based on variables including:

• fixed acidity
• volatile acidity
• citric acid
• residual sugar
• chlorides
• free sulfur dioxide
• total sulfur dioxide
• density
• pH
• sulphates
• alcohol

Moreover, we will also attempt to look for the correlation between the explanatory variables, to see if there are any multicollinearity occurs. Assuming all variables are independent, if the variables are correlated and the degree of correlation between variables is high enough, it can be a problem when fitting the model and the result may not be accurate.

Analysis

The k-nearest neighbors (k-nn) algorithm will be used to build a regression model to predict the quality score of the wine, including white and red wine. All variables included in the original dataset will be used in fitting our model. Since there are no missing values in our dataset, there is no need for any imputation. Also, the ridge regression model will be our second model in predicting the quality score of wine. We are hoping to see improvement in our model. Most of our analysis will be done using Python and R programming languages (R Core Team 2020; Van Rossum and Drake 2009).

Exploratory Data Analysis (EDA)

EDA table:

• We will be using python data.info() to check for information on our data. From this table, we can check for the data type of each column and look for missing values contained in every column. If there are any missing values, we would want to clean the data first before proceeding with any analysis. Also, data.describe is also useful in identifying the minimum, maximum, and mean of each numerical columns. We would want to double-check if the values returned from the table are ‘acceptable’.

EDA figure:

• We will be plotting a histogram on all the variables to gain insight into the probability distribution that the dataset follows. With a histogram, it is easy for us to see how the data is distributed. Also, scatterplot will be used to show the relationship between two variables. If the points between two variables resemble a straight line, this indicates that the relationship between the two variables we have is approximately linear.

Sharing Results of Analysis

We would compare the results of our cross-validation analysis to test analysis and see whether there exists any overfitting or underfitting in our model. Also, we would return the confusion matrix of model performance on the test data to check for misclassification errors.

Report

The final report can be found here.

Usage

To replicate the analysis, clone this GitHub repository, install the dependencies listed below, and run the following commands at the command line/terminal from the root directory of this project:

# download data
python src/Download_file.py --source=https://archive.ics.uci.edu/ml/machine-learning-databases/wine-quality/winequality-red.csv --destination=data/winequality-red.csv
python src/Download_file.py --source=https://archive.ics.uci.edu/ml/machine-learning-databases/wine-quality/winequality-white.csv --destination=data/winequality-white.csv
# pre-process data 
python src/Preprocessing.py --raw_white=data/winequality-white.csv --raw_red=data/winequality-red.csv --preprocessed_train=data/winequality-train.csv --preprocessed_test=data/winequality-test.csv
# create exploratory data visualizations
python src/EDA.py --preprocessed_train=data/winequality-train.csv --quality_count_path=results/quality_count.png --quality_all_variables_path=results/quality_all_variables.png
# tune and test model
python src/ML_analyses.py --preprocessed_train=data/winequality-train.csv --preprocessed_test=data/winequality-test.csv --results_path=results
# render final report
Rscript -e "rmarkdown::render('doc/wine_quality_predict_report.Rmd')"

Dependencies

• Python 3.8.3 and Python packages:

  • ipykernel
  • docopt==0.6.2
  • pandas==0.24.2
  • altair>=4.1.0
  • scikit-learn>=0.23.2

• R version 4.0.3 and R packages:

  • knitr==1.29
  • tidyverse==1.2.1
  • numpy==1.19.1
  • dplyr==1.0.2
  • readr==1.4.0

• GNU make 4.2.1

References

Cortez, Paulo, António Cerdeira, Fernando Almeida, Telmo Matos, and José Reis. 2009. “Modeling Wine Preferences by Data Mining from Physicochemical Properties.” Decision Support Systems 47 (4): 547–53.

Dua, Dheeru, and Casey Graff. 2017. “UCI Machine Learning Repository.” University of California, Irvine, School of Information; Computer Sciences. http://archive.ics.uci.edu/ml.

R Core Team. 2020. R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing. https://www.R-project.org/.

Van Rossum, Guido, and Fred L. Drake. 2009. Python 3 Reference Manual. Scotts Valley, CA: CreateSpace.