2019 5th International Conference on Education Technology, Management and Humanities Science (ETMHS 2019)

Identification of red wine categories based on physicochemical properties

Xueting Bai1, Lingbo Wang1, Hanning Li2
1
School of Statistics, Shanxi University of Finance and Economics, Taiyuan, China
2
Faculty of International Trade, Shanxi University of Finance and Economics, Taiyuan, China

Keywords: Portrait of wine; Factor analysis; Cluster analysis; SPSS

Abstract: This paper mainly carried out the analysis of red wine categories. The red wine dataset is
subjected to dimension reduction and clustering of samples through empirical analysis. First, factor
analysis is performed on the 13 variables, and the complex variables are classified into five types of
factors, namely the bitter trophic factor, the visual evaluation factor, the hue factor, the pH factor
and the mineral element factor. Second, the samples were clustered by K-mean cluster analysis, and
the samples were clustered into three different varieties. According to the cluster center, the
characteristics of each variety can be summarized. Through a series of empirical analyses, a rough
portrait of the red wine characteristics can be made and categories can be clustered in this data set.

1. Introduction
With the advancement of society and the continuous improvement of the life quality, the public
demand for wine is gradually increasing. The wine culture is gradually becoming the civilization of
all mankind. Thus people need more knowledge about wine to help them understand the wine
civilization. The various physical properties of red wine (such as color, brightness and darkness)
and the content of some of the ingredients can be used to quantitatively analyze the characteristics
of different varieties of red wine. Different varieties have obvious differences in physical and
chemical properties. Using the differences between the types reflected by these properties, even
people who are unfamiliar with red wine can easily identify them. Therefore, this paper will use the
various indicators of red wine to distinguish them and classify them.

2. Introduction to the method used by the model

2.1 Experimental data and variable interpretation
The data used in this experiment was obtained from the UCI database of the Italian wine data set,
which contained a sample size of 178. The data contained in each variable is the result of a
chemical analysis. The Italian wines shown in the sample are grown in the same area but from
different varieties. The data set consists of a total of 13 numeric variables.
(1) Malic acid: It is a kind of acid with strong acidity and apple aroma. The red wine is naturally
accompanied by malic acid. (2) Ash: The essence of ash is an inorganic salt, which has an effect on
the overall flavor of the wine and can give the wine a fresh feeling. (3) Alkalinity of ash: It is a
measure of weak alkalinity dissolved in water. (4) Magnesium: It is an essential element of the
human body, which can promote energy metabolism and is weakly alkaline. (5) Total phenols:
molecules containing polyphenolic substances, which have a bitter taste and affect the taste, color
and taste of the wine, and belong to the nutrients in the wine. (6) Flavanoids: It is a beneficial
antioxidant for the heart and anti-aging, rich in aroma and bitter. (7) Nonflavanoid phenols: It is a
special aromatic gas with oxidation resistance and is weakly acidic. (8) Proanthocyanins: It is a
bioflavonoid compound, which is also a natural antioxidant with a slight bitter smell. (9) Color
intensity: refers to the degree of color shade. It is used to measure the style of wine to be “light” or
“thick”. The color intensity is high, meanwhile the longer the wine and grape juice are in contact
during the wine making process, the thicker the taste. (10) Hue: refers to the vividness of the color
and the degree of warmth and coldness. It can be used to measure the variety and age of the wine.

Copyright © (2019) Francis Academic Press, UK 1443 DOI: 10.25236/etmhs.2019.309

Red wines with higher ages will have a yellow hue and increased transparency. Color intensity and
hue are important indicators for evaluating the quality of a wine's appearance. (11) Proline: It is the
main amino acid in red wine and an important part of the nutrition and flavor of wine. (12)
OD280/OD315 of diluted wines: This is a method for determining the protein concentration, which
can determine the protein content of various wines.
2.2 Model description applied to the data set
Factor and cluster analysis are applied to the known data set to determine the category to which
the wine sample belongs. First, factor analysis will be performed on the 13 variables. The common
factors will be extracted and the factor load matrix will be rotated, and the characteristics of the
common factor will be summarized and finally the factor score formula and the comprehensive
score of the sample on the common factor will be obtained. Then it will be saved as a new variable
and K-mean cluster analysis will be performed on the observations, and the category characteristics
of each class will be summarized according to the final cluster center on each factor.

3. Empirical analysis
3.1 Dimensionality analysis of physical and chemical variables
Tab.1 KMO and Bartlett’s Test
Measure 0.779
Sampling a sufficient degree of Approximate chi square 1317.181
Kaiser-Meyer-Olkin df 78
Bartlett's sphericity test Sig. 0
KMO test is performed on the data set from UCI, and the result is shown in Tab.1. The KMO
value of the test is 0.779, which is suitable for factor analysis. When using SPSS for factor analysis,
if the factor whose feature root is greater than 1, it can be retained by default. As a result, the
system extracts 3 factors, and the cumulative contribution of variance is 66.53%. In order to extract
more information from the variables, the model is artificially set, which means five factors are
extracted, and a total variance of 80% is explained after this adjustment. The obtained composition
matrix is shown in Tab.2.
Tab.2 Unrotated Load Matrix
Component Matrixa
1 2 3 4 5
Alcohol .313 .764 -.249 -.017 .245
Malic acid -.532 .355 .107 .515 -.033
Ash -.004 .499 .753 -.205 .132
Alkalinity of ash -.519 -.017 .736 .058 -.061
Magnesium .308 .473 .157 -.337 -.672
Total phenols .856 .103 .176 .190 .138
Flavanoids .917 -.005 .181 .146 .101
Nonflavanoid phenols -.648 .045 .205 -.195 .463
Proanthocyanins .680 .062 .180 .383 -.126
Color intensity -.192 .837 -.165 .063 .071
Hue .644 -.441 .102 -.410 .160
OD280/OD315 of diluted wines .816 -.260 .200 .177 .093
Proline .622 .577 -.152 -.222 .146
Observing the matrix before the factor rotation is performed, there is no significant difference in
the load of some common factors on the original variables. At this point, it is necessary to perform a
factor rotation on the load matrix.

1444
 Tab.3 Rotating Load Matrix
Rotated Component Matrixa
Component
1 2 3 4 5
Alcohol .148 .876 -.109 -.011 .058
Malic acid -.146 .006 -.789 .173 -.112
Ash .084 .243 -.032 .886 .155
Alkalinity of ash -.182 -.424 -.294 .721 -.008
Magnesium .098 .229 .053 .148 .907
Total phenols .836 .268 .240 .031 .026
Flavanoids .868 .189 .339 -.003 .051
Nonflavanoid phenols -.565 .004 -.075 .444 -.439
Proanthocyanins .795 .055 -.026 -.042 .155
Color intensity -.198 .696 -.472 .118 .122
Hue .352 -.087 .825 -.022 -.021
OD280/OD315 of diluted wines .816 -.058 .374 -.048 -.046
Proline .324 .775 .245 -.007 .217
The factor rotation is performed on the load matrix, and the result is shown in the load matrix
after rotation in Tab.3. It is considered that the load is larger when the factor load is greater than
65%. Availability from Tab.3 is concluded below.
The first common factor carries a large load on the variables of total phenols, flavonoids,
proanthocyanins and OD280/OD315 of diluted wines. It is known that chemical substances such as
phenolic compounds, flavonoids, and proanthocyanins have a great influence on mouthfeel, and
have a feeling of sputum and bitterness. At the same time, these chemicals have strong antioxidant
properties and are, in some ways, the chemicals required by the human body. The higher
absorbance ratio of OD280/OD315 indicates high protein purity. Therefore the first common factor
can be named the bitter trophic factor of wine.
The second common factor has a large load on the variables of alcohol, color intensity and
proline. Proline is known to be an amino acid that regulates the flavor of the wine. The color
intensity refers to the degree of lightness of the color, and the greater the intensity, the darker the
color. It reflects the nature of the grapes that make the wine. Therefore the second common factor
can be named as the visual evaluation factor of wine.
The third common factor has a larger load on the two variables of malic acid and hue. Malic acid
is known to be a natural acid that balances the sweetness of wine. Malic acid is commonly used in
the production of wines for lactic acid fermentation (MLF), in which lactic acid bacteria convert the
more acidic malic acid into less acidic lactic acid. In the process of MLF, the total acid decreases
and the pH rises, which causes the color of the grape to change from purple to blue, thus changing
the color tone of the wine. The hue refers to the vividness and warmth of the color of the wine. So
the third common factor is the hue factor of wine.
The fourth common factor is heavily loaded on the two variables of ash and alkalinity of ash. It
is known that ash in wine is an effective substance for neutralizing acidity, and is essentially an
inorganic salt. Therefore the fourth common factor can be named the pH factor of wine.
The fifth common factor has a large load on the concentration variable of magnesium. Due to the
loss of the original data set, the original mineral element variables have lost, so the magnesium
element can temporarily represent the mineral element. The fifth common factor can be named as
the mineral element factor of wine.

1445
 Tab.4 Factor Score Matrix
Component Score Coefficient Matrix
1 2 3 4 5
Alcohol -.014 .431 -.021 -.017 -.157
Malic acid .214 -.055 -.541 .002 -.109
Ash .023 .089 .114 .602 .033
Alkalinity of ash .086 -.239 -.114 .446 .061
Magnesium -.118 -.086 .057 .049 .883
Total phenols .287 .063 -.054 .081 -.149
Flavanoids .276 .023 -.003 .071 -.107
Nonflavanoid phenols -.179 .170 .203 .313 -.422
Proanthocyanins .365 -.113 -.288 -.022 .048
Color intensity -.041 .316 -.193 .000 -.010
Hue -.094 .020 .512 .122 -.052
OD280/OD315 of diluted wines .282 -.079 .003 .050 -.143
Proline -.056 .363 .183 .032 .009
Tab.4 is a matrix of factor score coefficients. This gives the equations of the factor score, where
𝑋𝑋𝑖𝑖 (i=1,2,3,…,13) is the value normalized by the original data.
Y1 =
−0.014 X 1 + 0.214 X 2 + 0.023 X 3 + 0.086 X 4 − 0.118 X 5 + 0.287 X 6 + 0.276 X 7
−0.179 X 8 + 0.365 X 9 − 0.041X 10 − 0.094 X 11 + 0.282 X 12 − 0.056 X 13
Y2 = 0.431X 1 − 0.055 X 2 + 0.089 X 3 − 0.239 X 4 − 0.086 X 5 + 0.063 X 6 + 0.023 X 7
+0.170 X 8 − 0.113 X 9 + 0.316 X 10 + 0.020 X 11 − 0.079 X 12 + 0.363 X 13
Y3 =−0.021X 1 − 0.541X 2 + 0.114 X 3 − 0.114 X 4 + 0.057 X 5 − 0.054 X 6 − 0.003 X 7
+0.203 X 8 − 0.288 X 9 − 0.193 X 10 + 0.512 X 11 + 0.003 X 12 + 0.183 X 13
Y4 =−0.017 X 1 + 0.002 X 2 + 0.602 X 3 + 0.446 X 4 − 0.049 X 5 + 0.081X 6 + 0.071X 7
−0.313 X 8 − 0.022 X 9 + 0.000 X 10 + 0.122 X 11 + 0.050 X 12 + 0.032 X 13
Y5 =−0.157 X 1 − 0.109 X 2 + 0.033 X 3 + 0.061X 4 + 0.883 X 5 − 0.149 X 6 − 0.107 X 7
−0.422 X 8 + 0.048 X 9 − 0.010 X 10 − 0.052 X 11 − 0.143 X 12 + 0.009 X 13
Factor synthesis score of every Observations
0.26387 0.17424 0.15435 0.12096 0.08820
= ( Z1 + Z2 + Z3 + Z4 + Z5 )
0.80162 0.80162 0.80162 0.80162 0.80162
Z i ( i = 1, 2,3, 4,5 ) is the score of 178 data based on the 5 common factor.
3.2 Sample classification based on physical and chemical variables
Q-type cluster analysis was performed on the samples. The five factors obtained by factor
analysis were the bitter trophic factor, the visual evaluation factor, the hue factor, the pH factor and
the mineral element factor, which were saved as variables, and K-mean cluster analysis was
performed on the samples according to these five factors. At this point the data has been
standardized and there is no difference between the dimensions. Decide the number of clusters to be
3 categories.
Tab.5 Cluster Significance Test
Cluster Error
Mean df Mean df F Sig.
the bitter 24.602 2 0.73 175 33.69 0
the visual 14.288 2 0.848 175 16.846 0
the hue factor 17.358 2 0.813 175 21.349 0
the pH factor 2.375 2 0.984 175 2.413 0.093
the mineral 49.665 2 0.444 175 111.9 0

1446
 One-way ANOVA was performed on the differences between the three categories of factors
involved in the classification (the more significant the variance analysis has more influence on the
clustering results), the test results are shown in Tab. Only the P value of the pH factor is greater
than 0.05, indicating that the clustering results are generally effective.
Tab.6 Number of Clustering Cases
Number of cases in each cluster
1 40
cluster 2 63
3 75
valid 178
missing 0
As shown in Tab.6, the table shows the number of cases in each cluster. It can be found that the
proportions are roughly equal, and there is no large difference between the number of cases
included in the class. Therefore, the clustering can be considered to be good.
Tab.7 Final Cluster Center
Cluster
1 2 3
the bitter trophic factor -0.14302 -0.60303 0.58283
the visual evaluation 0.13214 -0.52691 0.37213
the hue factor -0.18645 0.58697 -0.39362
the pH factor 0.3028 -0.10057 -0.07701
the mineral element factor 1.38697 -0.42797 -0.38022
The final cluster center is obtained from Tab.7, and the following results can be obtained.
The first category: the bitter trophic factor score is slightly lower than the average value of 0. It
can be considered that its nutrient composition is slightly lower, and the taste is sweeter. The
nutritional value of this type of wine is low; the visual evaluation factor score is slightly larger than
the average value. It can be considered that the variable score used for visual evaluation are
moderate, while the degree of wine is moderate, the color is deep, and the flavor is easier to
distinguish; the hue factor score is slightly lower than the mean. It indicates that the wine has low
color saturation; the high pH factor indicates that the wine is more alkaline; the mineral element
factor content is higher, indicating that the magnesium content in the wine is higher.
The second category: bitter taste nutrient content is lower and lower than the first category of
wine, and the taste is sweeter. The nutritional value of this type is very low; and the visual
evaluation factor has the lowest score among the three categories, with the low degree, the light
color, and the Indistinguishable flavor of wine; the color saturation is the highest, so the color is
bright; and the alkalinity of the wine is the lowest; the substance has the lowest magnesium content.
The third category: the bitter trophic factor has the highest score, that is, the bitter taste nutrient
content is the highest. The nutritional value of this kind of wine is very high; the visual evaluation
factor has the highest score. The flavor is easy to distinguish, while the color is deep and the degree
of wine is higher; the liquid color saturation is the lowest; the alkalinity of wine is lower; the
mineral magnesium content is lower than the average level of the three categories.

4. Conclusion
The model performs factor analysis on the variables to obtain the five types of factors, namely
the bitter trophic factor, the visual evaluation factor, the hue factor, the pH factor and the mineral
element factor. K-mean cluster analysis is performed on the samples using these five factors, and
the samples were classified into three types. The first type of red wine is a kind of wine with a
sweet taste and low nutritional value, among which the mineral element content is high and it is an
alkaline wine. The second type of wine has the lowest nutritional value and belongs to sweet wine

1447
with bright colors. The third type tastes bitter, with the highest nutritional value and the deepest
color.

References
[1] Generally accepted rating principles: A primer [J] .Jan Pieter Krahnen, Martin Weber. Journal of
Banking and Finance .2001(1)
[2] Huang H J, Qian C B, Feng Fan, Zhou.X.Z. Based on the improved K-means algorithm, the
grading method of red wine according to the physical and chemical indicators of wine grapes and
wine is studied.[J].Chinese market,2017(16):196-197.
[3] Ye S P, Chen H G, Liang K H. Mathematical model for quality evaluation of wine physical and
chemical indicators.[J].Anhui Agricultural Sciences, 2015, 43(12): 214-216.
[4] Liu Chan, Jiang Wei. Wine grading based on wine physical and chemical indicators [J]. Henan
Science and Technology, 2014(16):30-31.
[5] Ma Jian, Yuan J H. Research on wine quality evaluation based on physical and chemical
indicators [J]. Food industry technology, 2013, 34(18):137-140+143.
[6] Dong Ying, Cui R X. Quality evaluation of red wine based on factor analysis [J]. Journal of
Dalian Nationalities University, 2014, 16(03):284-288.
[7] Cheng Z R, Jiu D K, Wang Y W. Graded evaluation model for wine quality [J]. China High-tech
Zone, 2018(03):58.
[8] Xu S S, Tan Bing, Li Qian, He Ting. Wine quality evaluation [J]. Modern trade industry, 2018,
39(09):54-56.
[9] Yi Liu, Jiawen Peng, and Zhihao Yu. 2018. Big Data Platform Architecture under The
Background of Financial Technology: In The Insurance Industry As An Example. In Proceedings of
the 2018 International Conference on Big Data Engineering and Technology (BDET 2018). ACM,
New York, NY, USA, 31-35.

1448