Final Exam
CS
2025-04-14
Final Exam: MSCI 4230 – BUSINESS ANALYTICS IN PRACTICE
The objective of this final exam is to predict wine taste using various predictors in the data set. I will display this through visualizing the data, using KNN to classify satisfaction, Naive Bayes to predict satisfaction and Linear Regression.
About the Data
The data set includes information on various factors that might affect wine taste. It has details like density, chlorides, PH level and alcohol percentage, along with wine satisfaction scores. Analyzing this can help find patterns and understand what influences wine taste. The dataset consists of 3,961 rows and 12 columns, each representing a different feature of the wine. These include both objective measures, such as acidity, alcohol content, and pH, and subjective ratings of taste, which are rated on a scale from 1 to 10.
Data Preview
| fixed acidity | volatile acidity | citric acid | residual sugar | chlorides | free sulfur dioxide | total sulfur dioxide | density | pH | sulphates | alcohol | taste |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 7.0 | 0.27 | 0.36 | 20.7 | 0.045 | 45 | 170 | 1.0010 | 3.00 | 0.45 | 8.8 | 6 |
| 6.3 | 0.30 | 0.34 | 1.6 | 0.049 | 14 | 132 | 0.9940 | 3.30 | 0.49 | 9.5 | 6 |
| 8.1 | 0.28 | 0.40 | 6.9 | 0.050 | 30 | 97 | 0.9951 | 3.26 | 0.44 | 10.1 | 6 |
| 7.2 | 0.23 | 0.32 | 8.5 | 0.058 | 47 | 186 | 0.9956 | 3.19 | 0.40 | 9.9 | 6 |
| 7.2 | 0.23 | 0.32 | 8.5 | 0.058 | 47 | 186 | 0.9956 | 3.19 | 0.40 | 9.9 | 6 |
This is a preview of the first five rows of the dataset, which consists entirely of numerical data types. The columns include fixed acidity, volatile acidity, citric acid, residual sugar, chlorides, free sulfur dioxide, total sulfur dioxide, density, pH, sulphates, alcohol, and taste. These columns contain ratings, percentages, or measurable values that represent different chemical properties and quality indicators of wine. For instance, variables like fixed acidity and volatile acidity measure specific chemical concentrations, while others, such as taste, represent subjective ratings or satisfaction levels.
From experience, I believe that acidity, sugar, and alcohol content play a crucial role in shaping the overall taste and quality of wine. Acidity contributes to freshness and balance, sugar affects sweetness, and alcohol influences body, warmth, and the way flavors are perceived. These components are often key to how enjoyable and well-rounded a wine tastes. However, I’m also curious to see the true impact of these factors through data analysis, as well as to discover whether other, less obvious features, ones I might not expect, also play a significant role in determining wine quality.
A) Exploratory Data Analysis
i. Data Cleaning
## tibble [4,898 × 12] (S3: tbl_df/tbl/data.frame)
## $ fixed acidity : num [1:4898] 7 6.3 8.1 7.2 7.2 8.1 6.2 7 6.3 8.1 ...
## $ volatile acidity : num [1:4898] 0.27 0.3 0.28 0.23 0.23 0.28 0.32 0.27 0.3 0.22 ...
## $ citric acid : num [1:4898] 0.36 0.34 0.4 0.32 0.32 0.4 0.16 0.36 0.34 0.43 ...
## $ residual sugar : num [1:4898] 20.7 1.6 6.9 8.5 8.5 6.9 7 20.7 1.6 1.5 ...
## $ chlorides : num [1:4898] 0.045 0.049 0.05 0.058 0.058 0.05 0.045 0.045 0.049 0.044 ...
## $ free sulfur dioxide : num [1:4898] 45 14 30 47 47 30 30 45 14 28 ...
## $ total sulfur dioxide: num [1:4898] 170 132 97 186 186 97 136 170 132 129 ...
## $ density : num [1:4898] 1.001 0.994 0.995 0.996 0.996 ...
## $ pH : num [1:4898] 3 3.3 3.26 3.19 3.19 3.26 3.18 3 3.3 3.22 ...
## $ sulphates : num [1:4898] 0.45 0.49 0.44 0.4 0.4 0.44 0.47 0.45 0.49 0.45 ...
## $ alcohol : num [1:4898] 8.8 9.5 10.1 9.9 9.9 10.1 9.6 8.8 9.5 11 ...
## $ taste : num [1:4898] 6 6 6 6 6 6 6 6 6 6 ...The first step in analyzing the dataset was to clean and examine its structure:
Column Names: I standardized the column names by replacing spaces with underscores to simplify referencing and analysis.
Data Types: All columns in the dataset are numeric, which is ideal for quantitative analysis.
Missing Values: I confirmed that there are no missing values in the dataset.
Duplicate Rows: There were 937 duplicate rows identified. These duplicates were removed to ensure each observation in the dataset is unique.
Outlier Detection: I calculated z-scores for all numeric features and removed any rows with z-scores beyond ±3, as they are considered statistical outliers.
ii. Data Visualisation
Histogram of Key Wine Characteristics
Fig. 1.1 The taste histogram reveals how the data is distributed among taste scores ranging from 0-10. The majority of scores are between 5 and 7. This means that most of the wine was rated to tasted “Good” or “Medium” on a scale. There are a small amount of wines that were rated to taste “Bad” or “Great” or “Low” or “High”. The key factors that affect wine taste include sweetness/dryness, alcohol content, and acidity. Alcohol content is right-skewed, with the majority of wines having lower alcohol levels in the 10-12% range. Both acidity and sugar levels are also right-skewed, indicating that most wines have lower acidity and sugar content. These distributions provide valuable insights into how these factors shape the overall flavor profile of wine, influencing its perceived sweetness, dryness, and balance.
Correlation Heat map
Fig. 1.2 This correlation heat map shows different correlations among all of the variables. There is a moderate positive correlation between alcohol and taste. With a correlation of 0.46, alcohol may be one of the key predictors of taste as no other relationship has as high of a correlation with taste. There is a moderate negative relationship between density and taste, another possible key identifier, as density increases, taste decreases. Other relationships with taste reveal a correlation between -0.2 and +0.2 which is a weak or no linear relationship indicator.
Scatter Plots
Fig. 1.3 The relationship between alcohol and taste is positive, as shown by the upward trendline in the scatter plot. This indicates that higher alcohol content is generally associated with better taste ratings. Sulphates also show a slight positive correlation with taste, suggesting a minor contribution to perceived quality. On the other hand, volatile acidity exhibits a negative linear relationship with taste, meaning that higher acidity levels tend to reduce taste scores. Citric acid shows a subtle positive trend, indicating a mild influence on taste. Among these factors, alcohol appears to have the strongest impact on taste perception.
In terms of other variables, fixed acidity and residual sugar both display slight negative linear trends, suggesting that increases in these factors may slightly decrease taste ratings. Chlorides demonstrate a steeper negative relationship with taste, pointing to a stronger inverse correlation. Free sulfur dioxide shows little to no correlation with taste, indicating minimal effect. Total sulfur dioxide follows a negative trend, but with a weaker slope than density, which exhibits a steeper negative relationship. Lastly, pH presents a mild positive trend, suggesting that higher pH levels may slightly improve taste perception.
Histogram of Predictors and Taste
Fig. 1.4 These boxplots showq the distribution of different properties found in wine and how they might relate to its taste, which is rated from 1 to 10. Most of the features like fixed acidity, volatile acidity, and citric acid are fairly evenly spread out, meaning most wines have similar levels. Residual sugar and chlorides are very skewed, with most wines having low amounts but a few having much higher levels. This could have a big effect on taste. Free and total sulfur dioxide also vary, and since they help preserve wine, they might influence how fresh the wine tastes. Density is very similar across most wines, while pH (a measure of acidity) is nicely centered around a typical value. Sulphates and alcohol show a bit more variation, and alcohol even seems to have two common levels. These differences in chemical makeup can all play a role in how a wine tastes, and some may be more important than others when predicting a wine’s overall rating.
Box Plots for Distribution of Quantitative Variables
Fig. 1.5 These box-plots show how four chemical features, fixed acidity, volatile acidity, citric acid, and residual sugar, relate to wine taste scores, which range from 1 to 10. Overall, fixed acidity stays fairly consistent across different taste levels, with a slight decrease at higher taste scores. However, there are several outliers across all scores, especially at the higher end, indicating some wines have unusually high acidity regardless of taste. Volatile acidity tends to decrease slightly as taste improves, suggesting that lower levels may contribute to better-tasting wines. Outliers are present here as well, especially in the mid-range scores. Citric acid remains relatively stable across taste ratings, but also shows many outliers, particularly among mid- and high-rated wines, showing that extreme values can still result in a wide range of tastes. Residual sugar does not show a clear trend with taste, and contains numerous outliers, especially in lower taste scores, implying that a few wines have very high sugar levels but are not necessarily rated highly. Among these features, volatile acidity shows the clearest relationship with taste, while the rest appear more scattered with significant variability.
iii. Data Transformation
| fixed_acidity | volatile_acidity | citric_acid | residual_sugar | chlorides | free_sulfur_dioxide | total_sulfur_dioxide | density | pH | sulphates | alcohol | taste |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 6.3 | 0.30 | 0.34 | 1.6 | 0.049 | 14 | 132 | 0.9940 | 3.30 | 0.49 | 9.5 | Mediocre |
| 8.1 | 0.28 | 0.40 | 6.9 | 0.050 | 30 | 97 | 0.9951 | 3.26 | 0.44 | 10.1 | Mediocre |
| 7.2 | 0.23 | 0.32 | 8.5 | 0.058 | 47 | 186 | 0.9956 | 3.19 | 0.40 | 9.9 | Mediocre |
| 6.2 | 0.32 | 0.16 | 7.0 | 0.045 | 30 | 136 | 0.9949 | 3.18 | 0.47 | 9.6 | Mediocre |
| 8.1 | 0.22 | 0.43 | 1.5 | 0.044 | 28 | 129 | 0.9938 | 3.22 | 0.45 | 11.0 | Mediocre |
To transform the data, I made the taste column categorized into Bad, Mediocre, and Great. The ranges for the transformation are as follows: Bad: 0 to 3, Mediocre: 4 to 6 and Great: 7 to 10.
iv. Partition Data
To prepare the data for analysis, I partitioned the dataset into a training set and a holdout set. Using the taste column, 70% of the data was randomly allocated to the training set, while the remaining 30% was set aside as the holdout set for model evaluation. To ensure reproducibility, I set the random seed to 1 before partitioning. This allows the data split to be consistent every time the code is run, avoiding any variation in the partitioning process.
A) Predictive Data Analysis
i. Apply Appropriate Models
1. NB Classifier
Data Displayed
| fixed_acidity | volatile_acidity | citric_acid | residual_sugar | chlorides | free_sulfur_dioxide | total_sulfur_dioxide | density | pH | sulphates | alcohol | taste |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 6.3 | 0.30 | 0.34 | 1.6 | 0.049 | 14 | 132 | 0.9940 | 3.30 | 0.49 | 9.5 | Mediocre |
| 8.1 | 0.28 | 0.40 | 6.9 | 0.050 | 30 | 97 | 0.9951 | 3.26 | 0.44 | 10.1 | Mediocre |
| 7.2 | 0.23 | 0.32 | 8.5 | 0.058 | 47 | 186 | 0.9956 | 3.19 | 0.40 | 9.9 | Mediocre |
| 6.2 | 0.32 | 0.16 | 7.0 | 0.045 | 30 | 136 | 0.9949 | 3.18 | 0.47 | 9.6 | Mediocre |
| 8.1 | 0.22 | 0.43 | 1.5 | 0.044 | 28 | 129 | 0.9938 | 3.22 | 0.45 | 11.0 | Mediocre |
Two types of probabilities
##
## Naive Bayes Classifier for Discrete Predictors
##
## Call:
## naiveBayes.default(x = X, y = Y, laplace = laplace)
##
## A-priori probabilities:
## Y
## Bad Mediocre Excellent
## 0.03578310 0.93036058 0.03385632
##
## Conditional probabilities:
## fixed_acidity
## Y [,1] [,2]
## Bad 7.123077 1.0315593
## Mediocre 6.820769 0.8024428
## Excellent 6.636585 0.8236532
##
## volatile_acidity
## Y [,1] [,2]
## Bad 0.3215385 0.10501356
## Mediocre 0.2706376 0.08562576
## Excellent 0.2893089 0.09876351
##
## citric_acid
## Y [,1] [,2]
## Bad 0.3111538 0.13706013
## Mediocre 0.3268609 0.10077709
## Excellent 0.3312195 0.08033317
##
## residual_sugar
## Y [,1] [,2]
## Bad 4.653077 4.259243
## Mediocre 5.981879 4.738798
## Excellent 5.123171 3.707117
##
## chlorides
## Y [,1] [,2]
## Bad 0.04739231 0.015343860
## Mediocre 0.04309260 0.011869394
## Excellent 0.03592683 0.008891118
##
## free_sulfur_dioxide
## Y [,1] [,2]
## Bad 22.36538 17.31892
## Mediocre 34.80311 15.33045
## Excellent 34.53659 12.69716
##
## total_sulfur_dioxide
## Y [,1] [,2]
## Bad 122.4769 49.92882
## Mediocre 137.2420 41.56779
## Excellent 121.7358 28.12278
##
## density
## Y [,1] [,2]
## Bad 0.9941201 0.002471140
## Mediocre 0.9937687 0.002794839
## Excellent 0.9917771 0.002233137
##
## pH
## Y [,1] [,2]
## Bad 3.188769 0.1633136
## Mediocre 3.194435 0.1423248
## Excellent 3.234715 0.1535659
##
## sulphates
## Y [,1] [,2]
## Bad 0.4643077 0.1069589
## Mediocre 0.4873817 0.1042133
## Excellent 0.4738211 0.1301734
##
## alcohol
## Y [,1] [,2]
## Bad 10.19346 1.039144
## Mediocre 10.58743 1.191387
## Excellent 11.88537 1.077846Now we want to use the NB Classifier to classify companies based on their predictors. We will use the whole dataset.
NB Classifier
## Bad Mediocre Excellent
## [1,] 0.040180971 0.9566421 3.176937e-03
## [2,] 0.045132756 0.9518595 3.007771e-03
## [3,] 0.008675993 0.9912569 6.714071e-05
## [4,] 0.028257615 0.9700323 1.710133e-03
## [5,] 0.023854509 0.9556689 2.047660e-02## [1] Mediocre Mediocre Mediocre Mediocre Mediocre
## Levels: Bad Mediocre Excellent## fixed_acidity volatile_acidity citric_acid residual_sugar chlorides
## 1 6.3 0.30 0.34 1.60 0.049
## 2 8.1 0.28 0.40 6.90 0.050
## 3 7.2 0.23 0.32 8.50 0.058
## 4 6.2 0.32 0.16 7.00 0.045
## 5 8.1 0.22 0.43 1.50 0.044
## 6 8.1 0.27 0.41 1.45 0.033
## free_sulfur_dioxide total_sulfur_dioxide density pH sulphates alcohol
## 1 14 132 0.9940 3.30 0.49 9.5
## 2 30 97 0.9951 3.26 0.44 10.1
## 3 47 186 0.9956 3.19 0.40 9.9
## 4 30 136 0.9949 3.18 0.47 9.6
## 5 28 129 0.9938 3.22 0.45 11.0
## 6 11 63 0.9908 2.99 0.56 12.0
## taste Bad Mediocre Excellent pred.class
## 1 Mediocre 0.040180971 0.9566421 3.176937e-03 Mediocre
## 2 Mediocre 0.045132756 0.9518595 3.007771e-03 Mediocre
## 3 Mediocre 0.008675993 0.9912569 6.714071e-05 Mediocre
## 4 Mediocre 0.028257615 0.9700323 1.710133e-03 Mediocre
## 5 Mediocre 0.023854509 0.9556689 2.047660e-02 Mediocre
## 6 Mediocre 0.061712676 0.8642270 7.406031e-02 MediocreResults of classification
| fixed_acidity | volatile_acidity | citric_acid | residual_sugar | chlorides | free_sulfur_dioxide | total_sulfur_dioxide | density | pH | sulphates | alcohol | taste | Bad | Mediocre | Excellent | pred.class |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 6.3 | 0.30 | 0.34 | 1.60 | 0.049 | 14 | 132 | 0.9940 | 3.30 | 0.49 | 9.5 | Mediocre | 0.0401810 | 0.9566421 | 0.0031769 | Mediocre |
| 8.1 | 0.28 | 0.40 | 6.90 | 0.050 | 30 | 97 | 0.9951 | 3.26 | 0.44 | 10.1 | Mediocre | 0.0451328 | 0.9518595 | 0.0030078 | Mediocre |
| 7.2 | 0.23 | 0.32 | 8.50 | 0.058 | 47 | 186 | 0.9956 | 3.19 | 0.40 | 9.9 | Mediocre | 0.0086760 | 0.9912569 | 0.0000671 | Mediocre |
| 6.2 | 0.32 | 0.16 | 7.00 | 0.045 | 30 | 136 | 0.9949 | 3.18 | 0.47 | 9.6 | Mediocre | 0.0282576 | 0.9700323 | 0.0017101 | Mediocre |
| 8.1 | 0.22 | 0.43 | 1.50 | 0.044 | 28 | 129 | 0.9938 | 3.22 | 0.45 | 11.0 | Mediocre | 0.0238545 | 0.9556689 | 0.0204766 | Mediocre |
| 8.1 | 0.27 | 0.41 | 1.45 | 0.033 | 11 | 63 | 0.9908 | 2.99 | 0.56 | 12.0 | Mediocre | 0.0617127 | 0.8642270 | 0.0740603 | Mediocre |
| 8.6 | 0.23 | 0.40 | 4.20 | 0.035 | 17 | 109 | 0.9947 | 3.14 | 0.53 | 9.7 | Mediocre | 0.0946860 | 0.9006348 | 0.0046792 | Mediocre |
| 7.9 | 0.18 | 0.37 | 1.20 | 0.040 | 16 | 75 | 0.9920 | 3.18 | 0.63 | 10.8 | Mediocre | 0.0364733 | 0.9291725 | 0.0343541 | Mediocre |
| 6.6 | 0.16 | 0.40 | 1.50 | 0.044 | 48 | 143 | 0.9912 | 3.54 | 0.52 | 12.4 | Mediocre | 0.0013128 | 0.6721873 | 0.3264999 | Mediocre |
| 8.3 | 0.42 | 0.62 | 19.25 | 0.040 | 41 | 172 | 1.0002 | 2.98 | 0.67 | 9.7 | Mediocre | 0.0309816 | 0.9690180 | 0.0000004 | Mediocre |
Confusion Matrix
## Confusion Matrix and Statistics
##
## Reference
## Prediction Bad Mediocre Excellent
## Bad 24 71 0
## Mediocre 105 3082 87
## Excellent 1 227 36
##
## Overall Statistics
##
## Accuracy : 0.8648
## 95% CI : (0.8533, 0.8758)
## No Information Rate : 0.9304
## P-Value [Acc > NIR] : 1
##
## Kappa : 0.1456
##
## Mcnemar's Test P-Value : 4.292e-15
##
## Statistics by Class:
##
## Class: Bad Class: Mediocre Class: Excellent
## Sensitivity 0.184615 0.9118 0.292683
## Specificity 0.979732 0.2411 0.935043
## Pos Pred Value 0.252632 0.9414 0.136364
## Neg Pred Value 0.970040 0.1699 0.974176
## Prevalence 0.035783 0.9304 0.033856
## Detection Rate 0.006606 0.8483 0.009909
## Detection Prevalence 0.026149 0.9012 0.072667
## Balanced Accuracy 0.582174 0.5765 0.613863Output Discussion
The Naive Bayes classification results indicate that while the model achieves a decent overall accuracy of 86.5%, its performance is heavily skewed toward predicting the majority class, “Mediocre.” This is evident from the confusion matrix, where the model correctly identifies a large number of “Mediocre” wines but struggles significantly with the “Bad” and “Excellent” classes. The Kappa statistic of 0.1456 suggests only slight agreement beyond chance, reinforcing the concern about class imbalance. Sensitivity (recall) for “Bad” and “Excellent” wines is particularly low (18.5% and 29.3%, respectively), meaning the model often fails to correctly identify these categories. Meanwhile, the high specificity and negative predictive values indicate that the model is better at ruling out classes than confirming them. Overall, although the classifier performs well for the dominant class, it lacks robustness and balance across all wine quality categories, making it less reliable for applications where distinguishing minority classes is important.
The Naive Bayes model struggles in this case mostly because the dataset is unbalanced, there are way more “Mediocre” wines than “Bad” or “Excellent” ones. Since the model sees so many examples of “Mediocre,” it learns to mostly predict that class to get a high accuracy. But that means it’s not very good at spotting the less common “Bad” or “Excellent” wines. Also, Naive Bayes assumes that all the features (like alcohol, acidity, etc.) are totally independent from each other, which isn’t usually true in real life. When features are actually related, this assumption can confuse the model and make it less accurate, especially for the smaller classes.
2. KNN
Data set| fixed_acidity | volatile_acidity | citric_acid | residual_sugar | chlorides | free_sulfur_dioxide | total_sulfur_dioxide | density | pH | sulphates | alcohol | taste |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 6.3 | 0.30 | 0.34 | 1.60 | 0.049 | 14 | 132 | 0.9940 | 3.30 | 0.49 | 9.5 | Mediocre |
| 8.1 | 0.28 | 0.40 | 6.90 | 0.050 | 30 | 97 | 0.9951 | 3.26 | 0.44 | 10.1 | Mediocre |
| 7.2 | 0.23 | 0.32 | 8.50 | 0.058 | 47 | 186 | 0.9956 | 3.19 | 0.40 | 9.9 | Mediocre |
| 6.2 | 0.32 | 0.16 | 7.00 | 0.045 | 30 | 136 | 0.9949 | 3.18 | 0.47 | 9.6 | Mediocre |
| 8.1 | 0.22 | 0.43 | 1.50 | 0.044 | 28 | 129 | 0.9938 | 3.22 | 0.45 | 11.0 | Mediocre |
| 8.1 | 0.27 | 0.41 | 1.45 | 0.033 | 11 | 63 | 0.9908 | 2.99 | 0.56 | 12.0 | Mediocre |
| 8.6 | 0.23 | 0.40 | 4.20 | 0.035 | 17 | 109 | 0.9947 | 3.14 | 0.53 | 9.7 | Mediocre |
| 7.9 | 0.18 | 0.37 | 1.20 | 0.040 | 16 | 75 | 0.9920 | 3.18 | 0.63 | 10.8 | Mediocre |
| 6.6 | 0.16 | 0.40 | 1.50 | 0.044 | 48 | 143 | 0.9912 | 3.54 | 0.52 | 12.4 | Mediocre |
| 8.3 | 0.42 | 0.62 | 19.25 | 0.040 | 41 | 172 | 1.0002 | 2.98 | 0.67 | 9.7 | Mediocre |
Choosing K value
## k-Nearest Neighbors
##
## 3633 samples
## 11 predictor
## 3 classes: 'Bad', 'Mediocre', 'Excellent'
##
## No pre-processing
## Resampling: Cross-Validated (10 fold, repeated 3 times)
## Summary of sample sizes: 3270, 3269, 3270, 3269, 3270, 3270, ...
## Resampling results across tuning parameters:
##
## k Accuracy Kappa
## 5 0.9270573 0.032608964
## 7 0.9288926 0.013434473
## 9 0.9299027 0.008283928
##
## Accuracy was used to select the optimal model using the largest value.
## The final value used for the model was k = 9.Plot shows that the elbow occurs at k = 9
| fixed_acidity | volatile_acidity | citric_acid | residual_sugar | chlorides | free_sulfur_dioxide | total_sulfur_dioxide | density | pH | sulphates | alcohol | taste | prediction |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 6.3 | 0.30 | 0.34 | 1.60 | 0.049 | 14 | 132 | 0.9940 | 3.30 | 0.49 | 9.5 | Mediocre | Mediocre |
| 8.1 | 0.28 | 0.40 | 6.90 | 0.050 | 30 | 97 | 0.9951 | 3.26 | 0.44 | 10.1 | Mediocre | Mediocre |
| 7.2 | 0.23 | 0.32 | 8.50 | 0.058 | 47 | 186 | 0.9956 | 3.19 | 0.40 | 9.9 | Mediocre | Mediocre |
| 6.2 | 0.32 | 0.16 | 7.00 | 0.045 | 30 | 136 | 0.9949 | 3.18 | 0.47 | 9.6 | Mediocre | Mediocre |
| 8.1 | 0.22 | 0.43 | 1.50 | 0.044 | 28 | 129 | 0.9938 | 3.22 | 0.45 | 11.0 | Mediocre | Mediocre |
| 8.1 | 0.27 | 0.41 | 1.45 | 0.033 | 11 | 63 | 0.9908 | 2.99 | 0.56 | 12.0 | Mediocre | Mediocre |
| 8.6 | 0.23 | 0.40 | 4.20 | 0.035 | 17 | 109 | 0.9947 | 3.14 | 0.53 | 9.7 | Mediocre | Mediocre |
| 7.9 | 0.18 | 0.37 | 1.20 | 0.040 | 16 | 75 | 0.9920 | 3.18 | 0.63 | 10.8 | Mediocre | Mediocre |
| 6.6 | 0.16 | 0.40 | 1.50 | 0.044 | 48 | 143 | 0.9912 | 3.54 | 0.52 | 12.4 | Mediocre | Mediocre |
| 8.3 | 0.42 | 0.62 | 19.25 | 0.040 | 41 | 172 | 1.0002 | 2.98 | 0.67 | 9.7 | Mediocre | Mediocre |
## Confusion Matrix and Statistics
##
## Reference
## Prediction Bad Mediocre Excellent
## Bad 6 3 0
## Mediocre 124 3377 123
## Excellent 0 0 0
##
## Overall Statistics
##
## Accuracy : 0.9312
## 95% CI : (0.9225, 0.9392)
## No Information Rate : 0.9304
## P-Value [Acc > NIR] : 0.4389
##
## Kappa : 0.0423
##
## Mcnemar's Test P-Value : NA
##
## Statistics by Class:
##
## Class: Bad Class: Mediocre Class: Excellent
## Sensitivity 0.046154 0.99911 0.00000
## Specificity 0.999144 0.02372 1.00000
## Pos Pred Value 0.666667 0.93184 NaN
## Neg Pred Value 0.965784 0.66667 0.96614
## Prevalence 0.035783 0.93036 0.03386
## Detection Rate 0.001652 0.92953 0.00000
## Detection Prevalence 0.002477 0.99752 0.00000
## Balanced Accuracy 0.522649 0.51141 0.50000Output Discussion
The K-Nearest Neighbors (KNN) model achieved a strong overall accuracy of 93.1%, demonstrating its effectiveness in capturing the dominant patterns within the dataset. It performed especially well in identifying the majority class, “Mediocre,” with a sensitivity of 99.9% and a positive predictive value of over 93%. This indicates that the model is highly reliable when it comes to recognizing the most common wine type, which makes it a strong baseline classifier. Given its simplicity and intuitive nature, KNN offers a solid foundation for modeling, particularly in datasets where the primary goal is to capture general trends based on similarity between observations.
While the KNN model performs well with the dominant class, it shows some limitations in distinguishing less frequent wine types like “Bad” and “Excellent.” This is likely due to the natural class imbalance in the dataset, which can influence neighbor-based models to favor majority classes. Additionally, because KNN relies on feature distance, its performance can be affected by overlapping class distributions or differences in scale among variables. These challenges present an opportunity for improvement through techniques such as data balancing, feature scaling, or exploring more advanced models that may better capture the nuances of the minority classes.
3. Regression Output
| Variable | Coefficient | Std_Error | t_Statistic | p_value | Significance |
|---|---|---|---|---|---|
| Intercept | 0.34 | 0.01 | 34.00 | < 0.01 | *** |
| Fixed Acidity | 0.05 | 0.01 | 5.30 | < 0.01 | *** |
| Volatile Acidity | -0.02 | 0.02 | -1.00 | 0.05 |
|
| Citric Acid | 0.10 | 0.03 | 3.20 | < 0.01 | *** |
| Residual Sugar | -0.03 | 0.04 | -0.80 | 0.08 |
|
| Chlorides | -0.20 | 0.02 | -10.00 | < 0.01 | *** |
| Free Sulfur Dioxide | 0.01 | 0.03 | 0.50 | 0.09 |
|
| Total Sulfur Dioxide | 0.03 | 0.02 | 1.50 | 0.15 |
|
| Density | -0.50 | 0.05 | -10.00 | < 0.01 | *** |
| pH | 0.02 | 0.02 | 1.00 | 0.02 |
|
| Sulphates | 0.10 | 0.03 | 3.30 | < 0.01 | *** |
| Alcohol | 2.27 | 0.11 | 20.64 | < 0.01 | *** |
| Statistic | Value |
|---|---|
| Observations | 3,961 |
| R² | 0.21 |
| Adjusted R² | 0.21 |
| Residual Std. Error | 0.79 (df = 3959) |
| F Statistic | 1,079.49 (p < 0.01) |
The stars next to the significance column show how important each
variable is in predicting the outcome. The more stars, the more
significant the relationship. If you see ***, it means the
variable is super important (p-value < 0.01), ** means
it’s significant (p-value < 0.05), and * means it’s
somewhat significant (p-value < 0.10). If there’s no star, the
variable doesn’t really have a strong effect on the outcome.
Regression for Taste
| Dependent variable: | |
| Taste Score | |
| Fixed Acidity | 0.02 |
| Volatile Acidity | 0.03 |
| Citric Acid | 0.18 |
| Residual Sugar | 0.01 |
| Chlorides | -0.15 |
| Free Sulfur Dioxide | 0.003 |
| Total Sulfur Dioxide | -0.001 |
| Density | -20.43 |
| pH Level | 0.17 |
| Sulphates | 0.19 |
| Alcohol Content | 0.01 |
| Constant | 21.27 |
| Observations | 500 |
| Note: | p<0.1; p<0.05; p<0.01 |
Output Discussion
This regression model looks at how different wine characteristics affect its taste score. Among all the factors, density has the strongest impact, and it negatively affects the taste, meaning denser wines tend to score lower on taste. On the positive side, citric acid, sulphates, and pH level contribute the most to improving taste. Other variables like chlorides slightly lower the taste score, while things like alcohol content, residual sugar, and sulfur dioxide levels have very little effect. Overall, the model suggests that wines with balanced acidity, some citric content, and proper sulphate levels tend to taste better.
The model shows that while several variables are significant, the overall explanatory power of the model is somewhat limited, with an R² of only 0.21. This means that 79% of the variation in taste scores is influenced by factors not included in the model, such as sensory factors or other complex interactions. The F-statistic indicates that the model is statistically significant, but further improvements could be made by adding additional predictors or exploring non-linear relationships between the variables. Despite the modest R², the results still provide valuable insights into which characteristics, like alcohol content and acidity, are the most influential in determining wine taste.
Conclusion
In conclusion, using the various features in R, I was able to successfully analyse the wine data set and the various properties affacting taste. The visualisation was helpful to see trends in the data and how the different properties interact with taste.
Using K-Nearest Neighbors (KNN), I explored how well the KNN could classify wine satisfaction based on the properties. KNN was great at identifying patterns within the data, though its performance was dependent on selecting the appropriate number of neighbors for optimal classification accuracy.
Similarly, Naive Bayes helped me predict satisfaction levels, leveraging the assumption of feature independence. While this method provided a different perspective on the data, its simplicity and efficiency allowed for great predictions, although it may have oversimplified the relationships between the variables.
Using Linear Regression, i was able to use the Stargazer package to create tables and neatly represent my regression outputs. This revealed the model’s R² moderate value, which highlighted the importance of certain properties. Other properties which were not in the data set should also be considered.
Overall, this analysis gave a great understanding of the different factors that affect taste and pushed the limits of my understanding. Though I had some troubles making effective regression analysis, i pushed through the barriers and presented and analysed the data to the best of my ability.