This project analyses the various factors which can be used to classify wines into three classes. It studies the different properties across the 3 different classes, tries to find relationships amongst them and predicts the class from the property variants.
Initially we start by looking at the data present in the Dataset.
'data.frame': 178 obs. of 15 variables:
$ Al : num 11.6 12.6 14.1 14.1 13.8 ...
$ Mal : num 2.05 1.29 4.1 1.63 1.53 3.87 1.81 1.34 1.13 1.6 ...
$ Ash : num 3.23 2.1 2.74 2.28 2.7 2.4 2.41 1.9 2.16 2.52 ...
$ Alc : num 28.5 20 24.5 16 19.5 23 20.5 18.5 19 17.8 ...
$ Mg : int 119 103 96 126 132 101 100 88 87 95 ...
$ TotP : num 3.18 1.48 2.05 3 2.95 2.83 2.7 1.45 3.5 2.48 ...
$ Flav : num 5.08 0.58 0.76 3.17 2.74 2.55 2.98 1.36 3.1 2.37 ...
$ NFlav : num 0.47 0.53 0.56 0.24 0.5 0.43 0.26 0.29 0.19 0.26 ...
$ Pro : num 1.87 1.4 1.35 2.1 1.35 1.95 1.86 1.35 1.87 1.46 ...
$ Color : num 6 7.6 9.2 5.65 5.4 2.57 5.1 2.45 4.45 3.93 ...
$ Hue : num 0.93 0.58 0.61 1.09 1.25 1.19 1.04 1.04 1.22 1.09 ...
$ OD : num 3.69 1.55 1.6 3.71 3 3.13 3.47 2.77 2.87 3.63 ...
$ Prl : int 465 640 560 780 1235 463 920 562 420 1015 ...
$ Class : chr "two" "three" "three" "one" ...
There are 178 observations and 15 features. Input variables which includes 15 chemical and physical features of wine.
Below is brief description of each feature: Input variables (based on physicochemical tests):
Chemical and Physical Properties:
-
Alcohol: The percent alcohol content of the wine (% by volume)
-
Malic Acid : It is one of the principal organic acids found in wine grapes (g/l)
-
Ash : Ash content is one of the important indicators in wine quality determination (mS/cm)
-
Alcalinity : Several different types of acids found in wine affect how acidic a wine tastes. (pH)
-
Magnesium : Magnesium content in wines (gm per 1kg)
-
Total phenols : These are flavonoids that contribute to the construction of various tannins and contribute to the perception of bitterness in wine. (mg/L)
-
Flavanoid : Flavonoids are the most abundant polyphenols in the wine. (mg/L)
-
Non Flavanoid Phenols : Phenolic compounds in wine contribute specific characteristics to wine while also creating specific flavors and aromas when the complex interactions take place in wine during fermentation and wine-making. (mg/L)
-
Proanthocyanidins : A class of Phenol . (mg/L)
-
Color intensity : A simple measure of how dark the wine is.
-
Hue : It is one of the main properties of color.
-
OD280/ OD315 of Diluted Wine .
-
Proline : An amino-acid . (mg/L)
Output variable :
class ( classified into one, two and three )
Al Mal Ash Alc Mg
Min. :11.03 Min. :0.740 Min. :1.360 Min. :10.60 Min. : 70.00
1st Qu.:12.36 1st Qu.:1.603 1st Qu.:2.210 1st Qu.:17.20 1st Qu.: 88.00
Median :13.05 Median :1.865 Median :2.360 Median :19.50 Median : 98.00
Mean :13.00 Mean :2.336 Mean :2.367 Mean :19.49 Mean : 99.74
3rd Qu.:13.68 3rd Qu.:3.083 3rd Qu.:2.558 3rd Qu.:21.50 3rd Qu.:107.00
Max. :14.83 Max. :5.800 Max. :3.230 Max. :30.00 Max. :162.00
TotP Flav NFlav Pro Color
Min. :0.980 Min. :0.340 Min. :0.1300 Min. :0.410 Min. : 1.280
1st Qu.:1.742 1st Qu.:1.205 1st Qu.:0.2700 1st Qu.:1.250 1st Qu.: 3.220
Median :2.355 Median :2.135 Median :0.3400 Median :1.555 Median : 4.690
Mean :2.295 Mean :2.029 Mean :0.3619 Mean :1.591 Mean : 5.058
3rd Qu.:2.800 3rd Qu.:2.875 3rd Qu.:0.4375 3rd Qu.:1.950 3rd Qu.: 6.200
Max. :3.880 Max. :5.080 Max. :0.6600 Max. :3.580 Max. :13.000
Hue OD Prl Class class_num
Min. :0.4800 Min. :1.270 Min. : 278.0 Length:178 1:59
1st Qu.:0.7825 1st Qu.:1.938 1st Qu.: 500.5 Class :character 2:71
Median :0.9650 Median :2.780 Median : 673.5 Mode :character 3:48
Mean :0.9574 Mean :2.612 Mean : 746.9
3rd Qu.:1.1200 3rd Qu.:3.170 3rd Qu.: 985.0
Max. :1.7100 Max. :4.000 Max. :1680.0
Above figure shows the distribution of data over different variables. As we can see, the normal range for fixed acidity is 12.36 to 13.68 percent. As for malic acid, 75% of wines in our dataset have below 3.083 mg of malic acid in their content. Also, the Average alcohol percentage in our dataset is about 13.
Boxplot of Alcohol Content Percent :
For most of the wine in our dataset, alcohol percentage falls between 12 and 14.
Distribution of Acohol percentages and Color of intensity across the three classes is shown -
Alcohol Content :
Class 1 : 13 - 14.5 %
Class 2 : 11.5 - 13 %
Class 3 : 12 - 14.5 % .
We see there is a relation between the alcohol content and the class of wine .
Intensity of Color :
Class 1 have a moderate level of color ( 4-8 )
Class 2 seem to be lightly colored ( 1-6 )
Class 3 are darker in color ranging from 4 - 11 in intensity of color.
Distribution of Flavanoids and Malic Acid is shown -
Flavanoid Content appears to be moderately low in Class 1 wines ( 1-3 mg/L )
In class 2 wines it ranges from 0.5 - 4 mg/L .
In class 3 type of wines, it is spread from 1 - 6 mg/L .
We use Spearman's rho statistic to estimate a rank-based measure of association. Correlations falls between -1 and 1. 0 suggests there is no association between the two variables while numbers close to -1 or 1 suggests strong negative and positive associations accordingly.
> cor(wine[,1:13], wine$class_num)
[,1]
Al -0.33044357
Mal 0.43945626
Ash -0.04142315
Alc 0.51785911
Mg -0.20917939
TotP -0.71949973
Flav -0.84616171
NFlav 0.46064788
Pro -0.50040328
Color 0.26682296
Hue -0.60598251
OD -0.78604206
Prl -0.63371678
This shows that Wine Class has a positive correlation with Alcalinity and Malic Acid and negative correlation with OD280/OD315 , Flavanoids and Total Phenols .
The following diagram gives us a good sense of the distribution and correlation among input variables in our dataset:
A few of the observations from the above diagram -
-
Positive correlation between Flavanoids and Total Phenols
-
Positive correlation between OD280/OD315 and Total Phenols
-
Positive correlation between OD280/OD315 and Flavanoids
-
Positive correlation between Intensity of Color and Alcohol Content
-
Negative correlation between Non Flavanoid Phenols and Flavanoids
-
Negative correlation between OD280/OD315 and Non Flavanoid Phenols
To avoid multicollinearity in model building using regressions, we have to be aware of strong correlations among input variables.
The graph shows that low values of both variables are found in class 3 wines , moderately valued characteristics are displayed by class 2 wines and wines having high values of Flavanoids and OD belong to Class 1.
The graph shows that all wines show high total phenol content, however class 1 wines hve higher content than the rest of the classes
Class 3 type of wines have low Flavanoid levels while class 2 and class 1 have higher portions of Flavanoids in them.
We learn from this graph that Class 3 wines have low values of OD and Total Phenols which is also what we learnt from previous graphs.
However, Class 2 and Class 1 type of wines show higher levels of OD280/OD315 and Class 1 types of wines in particular have higher values of Total Phenols.
Below are Plottings of Total Phenol and OD280/OD315 faceted by the Class of the wines.
The results got are in tandem with individual plottings as well .
GBM Modeling is performed as there are 3 response variables in the Class.
The dataset is split into a 70:30 ratio. Training dataset occupies 70% of the dataset and the test dataset occupies 30 % of the dataset.
Predictions table got is as follows -
> table(test$class_num,Predict)
Predict
1 2 3
1 18 0 0
2 0 20 1
3 0 0 14
> acc
[1] 0.9811321
The summary of the Model provides the features and their weighted percentages :
> summary(model, order=TRUE)
var rel.inf
Prl Prl 26.49214259
Color Color 21.05442613
OD OD 19.61548943
Flav Flav 12.73613118
Al Al 10.45919260
Hue Hue 4.78965391
Alc Alc 2.45676213
Mal Mal 1.31343826
Ash Ash 0.39829753
Mg Mg 0.37031189
TotP TotP 0.27474601
Pro Pro 0.03940834
NFlav NFlav 0.00000000
We can see that Proline, Intensity of Color, OD280/OD315, Flavanoids and Alcohol Percenatage play a vital part in Predicting the Model.
Feature Selection is performed to find out which features affect the Predictions the most
The top predictors of Class type are found as follows -
> predictors(results)
[1] "Color" "Prl" "Flav" "OD" "Al" "Hue" "Mg" "TotP"
The following graph is a graph between OD280/OD315 and Proline and shows that the first class of wines can be easily predicted using the Proline values.
Proline values 800 and above have a high chance of being first class wine and 1000 and above are exclusive for first class wines.
Proline is therefore seen as the best predictor for the Class of wine .
