Report Wine Quality White

BI Project Report: Wine Quality - White

Bonaccini Valentina
16/09/2020

1. Abstract

"Wine Quality White" In the following report is reported the analysis carried out on the dataset with the aim of making a prediction on the preferences of wine consumers. To process the dataset was used a classification model: the Naïve Bayes. Applied in the world of wine producers, the analysis could help to increase the quality of the wines produced and support producers in their production and marketing choices.

2. Introduction and Background

Portugal is among the top 10 wine exporters, with 3.17% market share in 2005. The export of its Vinho Verde increased by 36% from 1997 to 2007. Nowadays, wine is becoming a drink for "Everyone" and we are increasingly working to develop technologies that help improve the production process and therefore the final quality of the wine. The development of new technologies in fact allows experts to collect and process large amounts of useful data to support the marketing process and to increase the company's "success" probability.

In particular, in the wine industry, decision support systems are mainly focused on the wine production phase, basing the analysis on the physicochemical data of the wines themselves. Vinho Verde White In this report, a case study is presented to model the consumer preferences based on its physicochemical data.

3. Dataset

3.1 Dataset Introduction

The study reported deals with a wine produced in Portugal. Vinho Verde represents the 15% of the total Portuguese production and about 10% of the total product is exported (mainly white). At the link https://archive.ics.uci.edu/ml/datasets/wine+quality there are two datasets: one for white wine and the other for red wine. The data was collected from May 2004 until February 2007. An important role in the collection was played by the CVRVV, an organization created with the aim of improving the quality of the wine produced and the marketing of Vinho Verde. Of interest for the analysis reported is the dataset concerning the "white" Vinho Verde. The dataset contains 4898 observations for each of which 11 physicochemical attributes (based on chemical tests) and a target (based on sensory data) are reported:

• Fixed acidity
• Volatile acidity
• Citric acid
• Residual sugar
• Chlorides
• Free solar dioxide
• Total sulfur dioxide
• Density
• pH
• Sulphates
• Alcohol
• Quality (target)

Analyzing in particular the target ("quality"), it represents the preferences of consumers. Each observation was evaluated by at least 3 testers who assigned a score from a minimum of 0 (very bad) to a maximum of 10 (excellent). The final value that is therefore found in "quality" is the median of testers' evaluations.

3.2 Data Exploration and Visualization

Starting from a first exploration of the data, it can certainly be useful to analyze the histogram of the "quality" target. As can be seen in figure 1 (frequency) and figure 2 (density), the distribution resembles a normal distribution (with more normal grades than external ones). In particular, in this case, calculating the kurtosis of the curve we have:

I = 0.2137674 ^kurt

Therefore we speak of a hypernormal: a distribution with less dispersion than the normal one.

Figure 1: histogram of wine quality (frequency)

Figure 2: histogram of wine quality (density)

For each of the remaining 11 physicochemical attributes, some preliminary statistics are reported in Table 1: mean, median, minimum value, and maximum value.

Table 1: statistics of physicochemical data of white wine

As we can see, the attributes have different ranges and different units of measure. For this reason, in paragraph 4.1.1, they will be normalized. For each of the 11 attributes, Figure 5 (page 5) reports the box plots after the normalization phase to have a more comprehensible general view.

A bivariate analysis was also developed on the data. In particular, the box plots between each attribute and the target were analyzed to compare distributions and are reported below.

4. Execution and Methods Used

The following paragraphs describe the procedures regarding the prep