Modeling wine preferences by data mining

from physicochemical properties

Paulo Cortez a,∗ António Cerdeira b Fernando Almeida b

Telmo Matos b José Reis a,b

a Department of Information Systems/R&D Centre Algoritmi, University of

Minho, 4800-058 Guimarães, Portugal

b Viticulture Commission of the Vinho Verde region (CVRVV), 4050-501 Porto,

Portugal

Abstract

We propose a data mining approach to predict human wine taste preferences that

is based on easily available analytical tests at the certification step. A large dataset

(when compared to other studies in this domain) is considered, with white and red

vinho verde samples (from Portugal). Three regression techniques were applied, un-

der a computationally efficient procedure that performs simultaneous variable and

model selection. The support vector machine achieved promising results, outper-

forming the multiple regression and neural network methods. Such model is useful

to support the oenologist wine tasting evaluations and improve wine production.

Furthermore, similar techniques can help in target marketing by modeling consumer

tastes from niche markets.

Key words: Sensory preferences, Regression, Variable selection, Model selection,

Support vector machines, Neural networks

Preprint submitted to Elsevier 22 May 2009

Once viewed as a luxury good, nowadays wine is increasingly enjoyed by a

wider range of consumers. Portugal is a top ten wine exporting country with

3.17% of the market share in 2005 [11]. Exports of its vinho verde wine (from

the northwest region) have increased by 36% from 1997 to 2007 [8]. To support

its growth, the wine industry is investing in new technologies for both wine

making and selling processes. Wine certification and quality assessment are

key elements within this context. Certification prevents the illegal adulteration

of wines (to safeguard human health) and assures quality for the wine market.

Quality evaluation is often part of the certification process and can be used

to improve wine making (by identifying the most influential factors) and to

stratify wines such as premium brands (useful for setting prices).

Wine certification is generally assessed by physicochemical and sensory tests

[10]. Physicochemical laboratory tests routinely used to characterize wine in-

clude determination of density, alcohol or pH values, while sensory tests rely

mainly on human experts. It should be stressed that taste is the least un-

derstood of the human senses [25], thus wine classification is a difficult task.

Moreover, the relationships between the physicochemical and sensory analysis

are complex and still not fully understood [20].

Advances in information technologies have made it possible to collect, store

and process massive, often highly complex datasets. All this data hold valu-

able information such as trends and patterns, which can be used to improve

∗ Corresponding author. E-mail pcortez@dsi.uminho.pt; tel.: +351 253510313; fax:

+351 253510300.

2
decision making and optimize chances of success [28]. Data mining (DM) tech-

niques [33] aim at extracting high-level knowledge from raw data. There are

several DM algorithms, each one with its own advantages. When modeling con-

tinuous data, the linear/multiple regression (MR) is the classic approach. The

backpropagation algorithm was first introduced in 1974 [32] and later popular-

ized in 1986 [23]. Since then, neural networks (NNs) have become increasingly

used. More recently, support vector machines (SVMs) have also been proposed

[4][26]. Due to their higher flexibility and nonlinear learning capabilities, both

NNs and SVMs are gaining an attention within the DM field, often attaining

high predictive performances [16][17]. SVMs present theoretical advantages

over NNs, such as the absence of local minima in the learning phase. In effect,

the SVM was recently considered one of the most influential DM algorithms

[34]. While the MR model is easier to interpret, it is still possible to extract

knowledge from NNs and SVMs, given in terms of input variable importance

[18][7].

When applying these DM methods, variable and model selection are critical

issues. Variable selection [14] is useful to discard irrelevant inputs, leading

to simpler models that are easier to interpret and that usually give better

performances. Complex models may overfit the data, losing the capability

to generalize, while a model that is too simple will present limited learning

capabilities. Indeed, both NN and SVM have hyperparameters that need to

be adjusted [16], such as the number of NN hidden nodes or the SVM kernel

parameter, in order to get good predictive accuracy (see Section 2.3).

The use of decision support systems by the wine industry is mainly focused

on the wine production phase [12]. Despite the potential of DM techniques to

predict wine quality based on physicochemical data, their use is rather scarce

3
and mostly considers small datasets. For example, in 1991 the “Wine” dataset

was donated into the UCI repository [1]. The data contain 178 examples with

measurements of 13 chemical constituents (e.g. alcohol, Mg) and the goal is

to classify three cultivars from Italy. This dataset is very easy to discriminate

and has been mainly used as a benchmark for new DM classifiers. In 1997 [27],

a NN fed with 15 input variables (e.g. Zn and Mg levels) was used to predict

six geographic wine origins. The data included 170 samples from Germany

and a 100% predictive rate was reported. In 2001 [30], NNs were used to

classify three sensory attributes (e.g. sweetness) of Californian wine, based

on grape maturity levels and chemical analysis (e.g. titrable acidity). Only

36 examples were used and a 6% error was achieved. Several physicochemical

parameters (e.g. alcohol, density) were used in [20] to characterize 56 samples

of Italian wine. Yet, the authors argued that mapping these parameters with a

sensory taste panel is a very difficult task and instead they used a NN fed with

data taken from an electronic tongue. More recently, mineral characterization

(e.g. Zn and Mg) was used to discriminate 54 samples into two red wine

classes [21]. A probabilistic NN was adopted, attaining 95% accuracy. As a

powerful learning tool, SVM has outperformed NN in several applications,

such as predicting meat preferences [7]. Yet, in the field of wine quality only

one application has been reported, where spectral measurements from 147

bottles were successfully used to predict 3 categories of rice wine age [35].

In this paper, we present a case study for modeling taste preferences based on

analytical data that are easily available at the wine certification step. Build-

ing such model is valuable not only for certification entities but also wine

producers and even consumers. It can be used to support the oenologist wine

evaluations, potentially improving the quality and speed of their decisions.

4
Moreover, measuring the impact of the physicochemical tests in the final wine

quality is useful for improving the production process. Furthermore, it can

help in target marketing [24], i.e. by applying similar techniques to model the

consumers preferences of niche and/or profitable markets.

The main contributions of this work are:

• We present a novel method that performs simultaneous variable and model

selection for NN and SVM techniques. The variable selection is based on

sensitivity analysis [18], which is a computationally efficient method that

measures input relevance and guides the variable selection process. Also, we

propose a parsimony search method to select the best SVM kernel parameter

with a low computational effort.

• We test such approach in a real-world application, the prediction of vinho

verde wine (from the Minho region of Portugal) taste preferences, showing

its impact in this domain. In contrast with previous studies, a large dataset

is considered, with a total of 4898 white and 1599 red samples. Wine pref-

erences are modeled under a regression approach, which preserves the order

of the grades, and we show how the definition of the tolerance concept is

useful for accessing different performance levels. We believe that this inte-

grated approach is valuable to support applications where ranked sensory

preferences are required, for example in wine or meat quality assurance.

The paper is organized as follows: Section 2 presents the wine data, DM mod-

els and variable selection approach; in Section 3, the experimental design is

described and the obtained results are analyzed; finally, conclusions are drawn

in Section 4.

5
2 Materials and methods

2.1 Wine data

This study will consider vinho verde, a unique product from the Minho (north-

west) region of Portugal. Medium in alcohol, is it particularly appreciated due

to its freshness (specially in the summer). This wine accounts for 15% of the

total Portuguese production [8], and around 10% is exported, mostly white

wine. In this work, we will analyze the two most common variants, white and

red (rosé is also produced), from the demarcated region of vinho verde. The

data were collected from May/2004 to February/2007 using only protected

designation of origin samples that were tested at the official certification en-

tity (CVRVV). The CVRVV is an inter-professional organization with the

goal of improving the quality and marketing of vinho verde. The data were

recorded by a computerized system (iLab), which automatically manages the

process of wine sample testing from producer requests to laboratory and sen-

sory analysis. Each entry denotes a given test (analytical or sensory) and the

final database was exported into a single sheet (.csv).

During the preprocessing stage, the database was transformed in order to

include a distinct wine sample (with all tests) per row. To avoid discarding

examples, only the most common physicochemical tests were selected. Since

the red and white tastes are quite different, the analysis will be performed

separately, thus two datasets 1 were built with 1599 red and 4898 white exam-

ples. Table 1 presents the physicochemical statistics per dataset. Regarding

the preferences, each sample was evaluated by a minimum of three sensory

1 The datasets are available at: http://www3.dsi.uminho.pt/pcortez/wine/

6
assessors (using blind tastes), which graded the wine in a scale that ranges

from 0 (very bad) to 10 (excellent). The final sensory score is given by the me-

dian of these evaluations. Fig. 1 plots the histograms of the target variables,

denoting a typical normal shape distribution (i.e. with more normal grades

that extreme ones).

[ insert Table 1 and Fig. 1 around here ]

2.2 Data mining approach and evaluation

We will adopt a regression approach, which preserves the order of the prefer-

ences. For instance, if the true grade is 3, then a model that predicts 4 is better

than one that predicts 7. A regression dataset D is made up of k ∈ {1, ..., N }

examples, each mapping an input vector with I input variables (xk1 , . . . , xkI ) to

a given target yk . The regression performance is commonly measured by an

error metric, such as the mean absolute deviation (MAD) [33]:

M AD =
PN
|yi − ybi |/N (1)
i=1

where ybk is the predicted value for the k input pattern. The regression error

characteristic (REC) curve [2] is also used to compare regression models, with

the ideal model presenting an area of 1.0. The curve plots the absolute error

tolerance T (x-axis), versus the percentage of points correctly predicted (the

accuracy) within the tolerance (y-axis).

The confusion matrix is often used for classification analysis, where a C × C

matrix (C is the number of classes) is created by matching the predicted

values (in columns) with the desired classes (in rows). For an ordered output,

7
the predicted class is given by pi = yi , if |yi − ybi | ≤ T , else pi = yi0 , where yi0

denotes the closest class to ybi , given that yi0 6= yi . From the matrix, several

metrics can be used to access the overall classification performance, such as

the accuracy and precision (i.e. the predicted column accuracies) [33].

The holdout validation is commonly used to estimate the generalization capa-

bility of a model [19]. This method randomly partitions the data into training

and test subsets. The former subset is used to fit the model (typically with 2/3

of the data), while the latter (with the remaining 1/3) is used to compute the

estimate. A more robust estimation procedure is the k-fold cross-validation

[9], where the data is divided into k partitions of equal size. One subset is

tested each time and the remaining data are used for fitting the model. The

process is repeated sequentially until all subsets have been tested. Therefore,

under this scheme, all data are used for training and testing. However, this

method requires around k times more computation, since k models are fitted.

2.3 Data mining methods

We will adopt the most common NN type, the multilayer perceptron, where

neurons are grouped into layers and connected by feedforward links [3]. For

regression tasks, this NN architecture is often based on one hidden layer of

H hidden nodes with a logistic activation and one output node with a linear

function [16]:
o−1
X 1
yb = wo,0 + PI · wo,i (2)
j=I+1 1 + exp(− i=1 xi wj,i − wj,0 )

where wi,j denotes the weight of the connection from node j to i and o the

output node. The performance is sensitive to the topology choice (H). A NN

8
with H = 0 is equivalent to the MR model. By increasing H, more complex

mappings can be performed, yet an excess value of H will overfit the data,

leading to generalization loss. A computationally efficient method to set H is

to search through the range {0, 1, 2, 3, . . . , Hmax } (i.e. from the simplest NN to

more complex ones). For each H value, a NN is trained and its generalization

estimate is measured (e.g. over a validation sample). The process is stopped

when the generalization decreases or when H reaches the maximum value

(Hmax ).

In SVM regression [26], the input x ∈ <I is transformed into a high m-

dimensional feature space, by using a nonlinear mapping (φ) that does not

need to be explicitly known but that depends of a kernel function (K). The

aim of a SVM is to find the best linear separating hyperplane, tolerating a

small error () when fitting the data, in the feature space:

m
X
yb = w0 + wi φi (x) (3)
i=1

The -insensitive loss function sets an insensitive tube around the residuals

and the tiny errors within the tube are discarded (Fig. 2).

[ insert Fig. 2 around here ]

We will adopt the popular gaussian kernel, which presents less parameters than

other kernels (e.g. polynomial) [31]: K(x, x0 ) = exp(−γ||x−x0 ||2 ), γ > 0. Under

this setup, the SVM performance is affected by three parameters: γ,  and C (a

trade-off between fitting the errors and the flatness of the mapping). To reduce

the search space, the first two values will be set using the heuristics [5]: C = 3
√
(for a standardized output) and  = σb / N , where σb = 1.5/N × N i=1 (yi − ybi )
2
P

and yb is the value predicted by a 3-nearest neighbor algorithm. The kernel

9
parameter (γ) produces the highest impact in the SVM performance, with

values that are too large or too small leading to poor predictions. A practical

method to set γ is to start the search from one of the extremes and then search

towards the middle of the range while the predictive estimate increases [31].

2.4 Variable and Model Selection

Sensitivity analysis [18] is a simple procedure that is applied after the train-

ing phase and analyzes the model responses when the inputs are changed.

Originally proposed for NNs, this sensitivity method can also be applied to

other algorithms, such as SVM [7]. Let ybaj denote the output obtained by

holding all input variables at their average values except xa , which varies

through its entire range with j ∈ {1, . . . , L} levels. If a given input variable

(xa ∈ {x1 , . . . , xI }) is relevant then it should produce a high variance (Va ).

Thus, its relative importance (Ra ) can be given by:

PL
Va = j=1 (ybaj − ybaj )2 /(L − 1)
(4)
PI
Ra = Va / i=1 Vi × 100 (%)

In this work, the Ra values will be used to measure the importance of the inputs

and also to discard irrelevant inputs, guiding the variable selection algorithm.

We will adopt the popular backward selection, which starts with all variables

and iteratively deletes one input until a stopping criterion is met [14]. Yet,

we guide the variable deletion (at each step) by the sensitivity analysis, in a

variant that allows a reduction of the computational effort by a factor of I

(when compared to the standard backward procedure) and that in [18] has

outperformed other methods (e.g. backward and genetic algorithms). Similarly

10
to [36], the variable and model selection will be performed simultaneously, i.e.

in each backward iteration several models are searched, with the one that

presents the best generalization estimate selected. For a given DM method,

the overall procedure is depicted bellow:

(1) Start with all F = {x1 , . . . , xI } input variables.

(2) If there is a hyperparameter P ∈ {P1 , . . . , Pk } to tune (e.g. NN or SVM),

start with P1 and go through the remaining range until the generalization

estimate decreases. Compute the generalization estimate of the model by

using an internal validation method. For instance, if the holdout method

is used, the available data are further split into training (to fit the model)

and validation sets (to get the predictive estimate).

(3) After fitting the model, compute the relative importances (Ri ) of all xi ∈

F variables and delete from F the least relevant input. Go to step 4 if

the stopping criterion is met, otherwise return to step 2.

(4) Select the best F (and P in case of NN or SVM) values, i.e., the input

variables and model that provide the best predictive estimates. Finally,

retrain this configuration with all available data.

3 Empirical results

The R environment [22] is an open source, multiple platform (e.g. Windows,

Linux) and high-level matrix programming language for statistical and data

analysis. All experiments reported in this work were written in R and con-

ducted in a Linux server, with an Intel dual core processor. In particular, we

adopted the RMiner [6], a library for the R tool that facilitates the use of

DM techniques in classification and regression tasks.

11
Before fitting the models, the data was first standardized to a zero mean and

one standard deviation [16]. RMiner uses the efficient BFGS algorithm to

train the NNs (nnet R package), while the SVM fit is based on the Sequential

Minimal Optimization implementation provided by LIBSVM (kernlab pack-

age). We adopted the default R suggestions [29]. The only exception are the

hyperparameters (H and γ), which will be set using the procedure described

in the previous section and with the search ranges of H ∈ {0, 1, . . . , 11} [36]

and γ ∈ {23 , 21 , . . . , 2−15 } [31]. While the maximum number of searches is

12/10, in practice the parsimony approach (step 2 of Section 2.4) will reduce

this number substantially.

Regarding the variable selection, we set the estimation metric to the M AD

value (Equation 1), as advised in [31]. To reduce the computational effort,

we adopted the simpler 2/3 and 1/3 holdout split as the internal valida-

tion method. The sensitivity analysis parameter was set to L = 5, i.e. xa ∈

{−1.0, −0.5, . . . , 1.0} for a standardized input. As a reasonable balance be-

tween the pressure towards simpler models and the increase of computational

search, the stopping criterion was set to 2 iterations without any improvement

or when only one input is available.

To evaluate the selected models, we adopted 20 runs of the more robust 5-fold

cross-validation, in a total of 20×5=100 experiments for each tested config-

uration. Statistical confidence will be given by the t-student test at the 95%

confidence level [13]. The results are summarized in Table 2. The test set

errors are shown in terms of the mean and confidence intervals. Three met-

rics are present: M AD, the classification accuracy for different tolerances (i.e.

T = 0.25, 0.5 and 1.0) and Kappa (T = 0.5). The selected models are described

in terms of the average number of inputs (I) and hyperparameter value (H or

12
γ). The last row shows the total computational time required in seconds.

[ insert Table 2 and Fig. 3 around here ]

For both tasks and all error metrics, the SVM is the best choice. The differences

are higher for small tolerances and in particular for the white wine (e.g. for

T = 0.25, the SVM accuracy is almost two times better when compared to

other methods). This effect is clearly visible when plotting the full REC curves

(Fig. 3). The Kappa statistic [33] measures the accuracy when compared with

a random classifier (which presents a Kappa value of 0%). The higher the

statistic, the more accurate the result. The most practical tolerance values are

T = 0.5 and T = 1.0. The former tolerance rounds the regression response

into the nearest class, while the latter accepts a response that is correct within

one of the two closest classes (e.g. a 3.1 value can be interpreted as grade 3

or 4 but not 2 or 5). For T = 0.5, the SVM accuracy improvement is 3.3

pp for red wine (6.2 pp for Kappa), a value that increases to 12.0 pp for the

white task (20.4 pp for Kappa). The NN is quite similar to MR in the red wine

modeling, thus similar performances were achieved. For the white data, a more

complex NN model (H = 2.1) was selected, slightly outperforming the MR

results. Regarding the variable selection, the average number of deleted inputs

ranges from 0.9 to 1.8, showing that most of the physicochemical tests used

are relevant. In terms of computational effort, the SVM is the most expensive

method, particularly for the larger white dataset.

A detailed analysis of the SVM classification results is presented by the average

confusion matrixes for T = 0.5 (Table 3). To simplify the visualization, the 3

and 9 grade predictions were omitted, since these were always empty. Most of

the values are close to the diagonals (in bold), denoting a good fit by the model.

13
The true predictive accuracy for each class is given by the precision metric

(e.g. for the grade 4 and white wine, precisionT =0.5 =19/(19+7+4)=63.3%).

This statistic is important in practice, since in a real deployment setting the

actual values are unknown and all predictions within a given column would

be treated the same. For a tolerance of 0.5, the SVM red wine accuracies

are around 57.7 to 67.5% in the intermediate grades (5 to 7) and very low

(0%/20%) for the extreme classes (3, 8 and 4), which are less frequent (Fig.

1). In general, the white data results are better: 60.3/63.3% for classes 6 and

4, 67.8/72.6% for grades 7 and 5, and a surprising 85.5% for the class 8 (the

exception are the 3 and 9 extremes with 0%, not shown in the table). When

the tolerance is increased (T = 1.0), high accuracies ranging from 81.9 to

100% are attained for both wine types and classes 4 to 8.

[ insert Table 3 and Fig. 4 around here ]

The average SVM relative importance plots (Ra values) of the analytical tests

are shown in Fig. 4. It should be noted that the whole 11 inputs are shown,

since in each simulation different sets of variables can be selected. In several

cases, the obtained results confirm the oenological theory. For instance, an

increase in the alcohol (4th and 2nd most relevant factor) tends to result in

a higher quality wine. Also, the rankings are different within each wine type.

For instance, the citric acid and residual sugar levels are more important in

white wine, where the equilibrium between the freshness and sweet taste is

more appreciated. Moreover, the volatile acidity has a negative impact, since

acetic acid is the key ingredient in vinegar. The most intriguing result is the

high importance of sulphates, ranked first for both cases. Oenologically this

result could be very interesting. An increase in sulphates might be related to

the fermenting nutrition, which is very important to improve the wine aroma.

14
4 Conclusions and implications

In recent years, the interest in wine has increased, leading to growth of the

wine industry. As a consequence, companies are investing in new technolo-

gies to improve wine production and selling. Quality certification is a crucial

step for both processes and is currently largely dependent on wine tasting by

human experts. This work aims at the prediction of wine preferences from

objective analytical tests that are available at the certification step. A large

dataset (with 4898 white and 1599 red entries) was considered, including vinho

verde samples from the northwest region of Portugal. This case study was ad-

dressed by two regression tasks, where each wine type preference is modeled

in a continuous scale, from 0 (very bad) to 10 (excellent). This approach pre-

serves the order of the classes, allowing the evaluation of distinct accuracies,

according to the degree of error tolerance (T ) that is accepted.

Due to advances in the data mining (DM) field, it is possible to extract knowl-

edge from raw data. Indeed, powerful techniques such as neural networks

(NNs) and more recently support vector machines (SVMs) are emerging. While

being more flexible models (i.e. no a priori restriction is imposed), the per-

formance depends on a correct setting of hyperparameters (e.g. number of

hidden nodes of the NN architecture or SVM kernel parameter). On the other

hand, the multiple regression (MR) is easier to interpret than NN/SVM, with

most of the NN/SVM applications considering their models as black boxes.

Another relevant aspect is variable selection, which leads to simpler models

while often improving the predictive performance. In this study, we present an

integrated and computationally efficient approach to deal with these issues.

Sensitivity analysis is used to extract knowledge from the NN/SVM models,

15
given in terms of relative importance of the inputs. Simultaneous variable and

model selection scheme is also proposed, where the variable selection is guided

by sensitivity analysis and the model selection is based on parsimony search

that starts from a reasonable value and is stopped when the generalization

estimate decreases.

Encouraging results were achieved, with the SVM model providing the best

performances, outperforming the NN and MR techniques, particularly for

white vinho verde wine, which is the most common type. When admitting

only the correct classified classes (T = 0.5), the overall accuracies are 62.4%

(red) and 64.6% (white). It should be noted that the datasets contain six/seven

classes (from 3 to 8/9). These accuracies are much better than the ones ex-

pected by a random classifier. The performance is substantially improved when

the tolerance is set to accept responses that are correct within the one of the

two nearest classes (T = 1.0), obtaining a global accuracy of 89.0% (red) and

86.8% (white). In particular, for both tasks the majority of the classes present

an individual accuracy (precision) higher than 90%.

The superiority of SVM over NN is probably due to the differences in the train-

ing phase. The SVM algorithm guarantees an optimum fit, while NN training

may fall into a local minimum. Also, the SVM cost function (Fig. 2) gives a

linear penalty to large errors. In contrast, the NN algorithm minimizes the sum

of squared errors. Thus, the SVM is expected to be less sensitive to outliers

and this effect results in a higher accuracy for low error tolerances. As argued

in [15], it is difficult to compare DM methods in a fair way, with data analysts

tending to favor models that they know better. We adopted the default sug-

gestions of the R tool [29], except for the hyperparameters (which were set

using a grid search). Since the default settings are more commonly used, this

16
seems a reasonable assumption for the comparison. Nevertheless, different NN

results could be achieved if different hidden node and/or minimization cost

functions were used. Under the tested setup, the SVM algorithm provided the

best results while requiring more computation. Yet, the SVM fitting can still

be achieved within a reasonable time with current processors. For example,

one run of the 5-fold cross-validation testing takes around 26 minutes for the

larger white dataset, which covers a three-year collection period.

The result of this work is important for the wine industry. At the certification

phase and by Portuguese law, the sensory analysis has to be performed by hu-

man tasters. Yet, the evaluations are based in the experience and knowledge of

the experts, which are prone to subjective factors. The proposed data-driven

approach is based on objective tests and thus it can be integrated into a

decision support system, aiding the speed and quality of the oenologist per-

formance. For instance, the expert could repeat the tasting only if her/his

grade is far from the one predicted by the DM model. In effect, within this

domain the T = 1.0 distance is accepted as a good quality control process and,

as shown in this study, high accuracies were achieved for this tolerance. The

model could also be used to improve the training of oenology students. Fur-

thermore, the relative importance of the inputs brought interesting insights

regarding the impact of the analytical tests. Since some variables can be con-

trolled in the production process this information can be used to improve the

wine quality. For instance, alcohol concentration can be increased or decreased

by monitoring the grape sugar concentration prior to the harvest. Also, the

residual sugar in wine could be raised by suspending the sugar fermentation

carried out by yeasts. Moreover, the volatile acidity produced during the malo-

lactic fermentation in red wine depends on the lactic bacteria control activity.

17
Another interesting application is target marketing [24]. Specific consumer

preferences from niche and/or profitable markets (e.g. for a particular coun-

try) could be measured during promotion campaigns (e.g. free wine tastings

at supermarkets) and modeled using similar DM techniques, aiming at the

design of brands that match these market needs.

Acknowledgments

We would like to thank Cristina Lagido and the anonymous reviewers for their

helpful comments. The work of P. Cortez is supported by the FCT project

PTDC/EIA/64541/2006.

References

[1] A. Asuncion and D. Newman. UCI Machine Learning Repository, University of

California, Irvine, http://www.ics.uci.edu/∼mlearn/MLRepository.html, 2007.

[2] J. Bi and K. Bennett. Regression Error Characteristic curves. In Proceedings

of 20th Int. Conf. on Machine Learning (ICML), Washington DC, USA, 2003.

[3] C. Bishop. Neural Networks for Pattern Recognition. Oxford University Press,

1995.

[4] B. Boser, I. Guyon, and V. Vapnik. A training algorithm for optimal margin

classifiers. In COLT ’92: Proceedings of the Fifth Annual Workshop on

Computational Learning Theory, pages 144–152, NY, USA, 1992. ACM.

[5] V. Cherkassy and Y. Ma. Practical Selection of SVM Parameters and Noise

Estimation for SVM Regression. Neural Networks, 17(1):113–126, 2004.

18
[6] P. Cortez. RMiner: Data Mining with Neural Networks and Support Vector

Machines using R. In R. Rajesh (Ed.), Introduction to Advanced Scientific

Softwares and Toolboxes, In press.

[7] P. Cortez, M. Portelinha, S. Rodrigues, V. Cadavez, and A. Teixeira. Lamb

Meat Quality Assessment by Support Vector Machines. Neural Processing

Letters, 24(1):41–51, 2006.

[8] CVRVV. Portuguese Wine - Vinho Verde. Comissão de Viticultura da Região

dos Vinhos Verdes (CVRVV), http://www.vinhoverde.pt, July 2008.

[9] T. Dietterich. Approximate Statistical Tests for Comparing Supervised

Classification Learning Algorithms. Neural Computation, 10(7):1895–1923,

1998.

[10] S. Ebeler. Flavor Chemistry - Thirty Years of Progress, chapter Linking

flavour chemistry to sensory analysis of wine, pages 409–422. Kluwer Academic

Publishers, 1999.

[11] FAO. FAOSTAT - Food and Agriculture Organization agriculture trade domain

statistics. http://faostat.fao.org/site/535/DesktopDefault.aspx?PageID=535,

July 2008.

[12] J. Ferrer, A. MacCawley, S. Maturana, S. Toloza, and J. Vera. An optimization

approach for scheduling wine grape harvest operations. International Journal

of Production Economics, 112(2):985–999, 2008.

[13] A. Flexer. Statistical evaluation of neural networks experiments: Minimum

requirements and current practice. In Proceedings of the 13th European Meeting

on Cybernetics and Systems Research, volume 2, pages 1005–1008, Vienna,

Austria, 1996.

[14] I. Guyon and A. Elisseeff. An introduction to variable and feature selection.

Journal of Machine Learning Research, 3(7–8):1157–1182, 2003.

19
[15] D. Hand. Classifier technology and the illusion of progress. Statistical Science,

21(1):1–15, 2006.

[16] T. Hastie, R. Tibshirani, and J. Friedman. The Elements of Statistical Learning:

Data Mining, Inference and Prediction. Springer-Verlag, NY, USA, 2001.

[17] Z. Huang, H. Chen, C. Hsu, W. Chen, and S. Wu. Credit rating analysis with

support vector machines and neural networks: a market comparative study.

Decision Support Systems, 37(4):543–558, 2004.

[18] R. Kewley, M. Embrechts, and C. Breneman. Data Strip Mining for the Virtual

Design of Pharmaceuticals with Neural Networks. IEEE Transactions on Neural

Networks, 11(3):668–679, May 2000.

[19] M. Kiang. A comparative assessment of classification methods. Decision

Support Systems, 35(4):441–454, 2003.

[20] A. Legin, A. Rudnitskaya, L. Luvova, Y. Vlasov, C. Natale, and A. D’Amico.

Evaluation of Italian wine by the electronic tongue: recognition, quantitative

analysis and correlation with human sensory perception. Analytica Chimica

Acta, 484(1):33–34, 2003.

[21] I. Moreno, D. González-Weller, V. Gutierrez, M. Marino, A. Cameán,

a. González, and A. Hardisson. Differentiation of two Canary DO red wines

according to their metal content from inductively coupled plasma optical

emission spectrometry and graphite furnace atomic absorption spectrometry

by using Probabilistic Neural Networks. Talanta, 72(1):263–268, 2007.

[22] R Development Core Team. R: A language and environment for statistical

computing. R Foundation for Statistical Computing, Vienna, Austria, ISBN

3-900051-00-3, http://www.R-project.org, 2008.

[23] D. Rumelhart, G. Hinton, and R. Williams. Learning Internal Representations

by Error Propagation. In D. Rulmelhart and J. McClelland, editors, Parallel

20
 Distributed Processing: Explorations in the Microstructures of Cognition,

volume 1, pages 318–362, MIT Press, Cambridge MA, 1986.

[24] M. Shaw, C. Subramaniam, G. Tan, and M. Welge. Knowledge management

and data mining for marketing. Decision Support Systems, 31(1):127–137, 2001.

[25] D. Smith and R. Margolskee. Making sense of taste. Scientific American,

Special issue, 16(3):84–92, 2006.

[26] A. Smola and B. Schölkopf. A tutorial on support vector regression. Statistics

and Computing, 14:199–222, 2004.

[27] L. Sun, K. Danzer, and G. Thiel. Classification of wine samples by means of

artificial neural networks and discrimination analytical methods. Fresenius’

Journal of Analytical Chemistry, 359(2):143–149, 1997.

[28] E. Turban, R. Sharda, J. Aronson, and D. King. Business Intelligence, A

Managerial Approach. Prentice-Hall, 2007.

[29] W. Venables and B. Ripley. Modern Applied Statistics with S. Springer, 4th

edition, 2003.

[30] S. Vlassides, J. Ferrier, and D. Block. Using Historical Data for Bioprocess

Optimization: Modeling Wine Characteristics Using Artificial Neural Networks

and Archived Process Information. Biotechnology and Bioengineering, 73(1),

2001.

[31] W. Wang, Z. Xu, W. Lu, and X. Zhang. Determination of the spread parameter

in the Gaussian kernel for classification and regression. Neurocomputing,

55(3):643–663, 2003.

[32] P. Werbos. Beyond regression: New tools for prediction and analysis in the

behavioral sciences. PhD thesis, Harvard University, Cambridge, MA, 1974.

21
[33] I.H. Witten and E. Frank. Data Mining: Practical Machine Learning Tools and

Techniques with Java Implementations. Morgan Kaufmann, San Francisco, CA,

2nd edition, 2005.

[34] X. Wu, V. Kumar, J. Quinlan, J. Gosh, Q. Yang, H. Motoda, G. MacLachlan,

A. Ng, B. Liu, P. Yu, Z. Zhou, M. Steinbach, D. Hand, and D. Steinberg. Top

10 algorithms in data mining. Knowledge and Information Systems, 14(1):1–37,

2008.

[35] H. Yu, H. Lin, H. Xu, Y. Ying, B. Li, and X. Pan. Prediction of Enological

Parameters and Discrimination of Rice Wine Age Using Least-Squares Support

Vector Machines and Near Infrared Spectroscopy. Agricultural and Food

Chemistry, 56(2):307–313, 2008.

[36] M. Yu, M. Shanker, G. Zhang, and M. Hung. Modeling consumer situational

choice of long distance communication with neural networks. Decision Support

Systems, 44(4):899–908, 2008.

22
Table 1

The physicochemical data statistics per wine type

Attribute (units) Red wine White wine

Min Max Mean Min Max Mean

fixed acidity (g(tartaric acid)/dm3 ) 4.6 15.9 8.3 3.8 14.2 6.9

volatile acidity (g(acetic acid)/dm3 ) 0.1 1.6 0.5 0.1 1.1 0.3

citric acid (g/dm3 ) 0.0 1.0 0.3 0.0 1.7 0.3

residual sugar (g/dm3 ) 0.9 15.5 2.5 0.6 65.8 6.4

chlorides (g(sodium chloride)/dm3 ) 0.01 0.61 0.08 0.01 0.35 0.05

free sulfur dioxide (mg/dm3 ) 1 72 14 2 289 35

total sulfur dioxide (mg/dm3 ) 6 289 46 9 440 138

density (g/cm3 ) 0.990 1.004 0.996 0.987 1.039 0.994

pH 2.7 4.0 3.3 2.7 3.8 3.1

sulphates (g(potassium sulphate)/dm3 ) 0.3 2.0 0.7 0.2 1.1 0.5

alcohol (% vol.) 8.4 14.9 10.4 8.0 14.2 10.4

23
Table 2

The wine modeling results (test set errors and selected models; best values in bold)

Red wine White wine

MR NN SVM MR NN SVM

MAD 0.50±0.00 0.51±0.00 0.46±0.00? 0.59±0.00 0.58±0.00 0.45±0.00?

AccuracyT =0.25 (%) 31.2±0.2 31.1±0.7 43.2±0.6? 25.6±0.1 26.5±0.3 50.3±1.1?

AccuracyT =0.50 (%) 59.1±0.1 59.1±0.3 62.4±0.4? 51.7±0.1 52.6±0.3 64.6±0.4?

AccuracyT =1.00 (%) 88.6±0.1 88.8±0.2 89.0±0.2♦ 84.3±0.1 84.7±0.1 86.8±0.2?

KappaT =0.5 (%) 32.2±0.3 32.5±0.6 38.7±0.7? 20.9±0.1 23.5±0.6 43.9±0.4?

Inputs (I) 9.2 9.3 9.8 9.6 9.3 10.1

Model – H=1 γ = 20.19 – H = 2.1 γ = 21.55

Time (s) 518 847 5589 551 1339 30674

♦ - Statistically significant under a pairwise comparison with MR.

24
Table 3

The average confusion matrixes (T = 0.5) and precision values (T = 0.5 and 1.0)

for the SVM model (bold values denote accurate predictions)

Actual Red wine predictions White wine predictions

Class 4 5 6 7 8 4 5 6 7 8

3 1 7 2 0 0 0 2 17 0 0

4 1 36 15 1 0 19 55 88 1 0

5 3 514 159 5 0 7 833 598 19 0

6 0 194 400 44 0 4 235 1812 144 3

7 0 10 107 82 1 0 18 414 441 7

8 0 0 10 8 0 0 3 71 43 59

9 0 1 3 2 0

PrecisionT =0.5 20.0% 67.5% 57.7% 58.6% 0.0% 63.3% 72.6% 60.3% 67.8% 85.5%

PrecisionT =1.0 93.8% 90.9% 86.6% 90.2% 100% 90.0% 93.3% 81.9% 90.3% 96.2%

25
 Red wine White wine

700

2000
Frequency (wine samples)

Frequency (wine samples)

1500
1000
300

500
100
0

0
3 4 5 6 7 8 3 4 5 6 7 8 9

Sensory preference Sensory preference

Fig. 1. The histograms for the red and white sensory preferences

26
 support +ε
vectors 0
−ε

−ε 0 +ε

Fig. 2. Example of a linear SVM regression and the -insensitive loss function

(adapted from [26])

27
 Red wine White wine

100

100
80

80
60

60
40

40
20

20
0

0
0 0.5 1 1.5 2 0 0.5 1 1.5 2

Fig. 3. The red (left) and white (right) wine average test set REC curves (SVM –

solid line, NN - gray line and MR – dashed line).

28
 Red wine

sulphates

pH

total sulfur dioxide

alcohol

volatile acidity

free sulfur dioxide

fixed acidity

residual sugar

chlorides

density

citric acid

0 5 10 15 20
White wine

sulphates

alcohol

residual sugar

citric acid

total sulfur dioxide

free sulfur dioxide

volatile acidity

density

pH

chlorides

fixed acidity

0 5 10 15 20

Fig. 4. The red (top) and white (bottom) wine input importances for the SVM

model (in %).

29