Analyzing and Predicting Alcohol or Non-Alcoholic Cocktails ^†

Abstract

Using a structured dataset, this study investigates the use of machine learning algorithms to analyze and forecast several properties of cocktails. Cocktails’ names, classifications, ingredients, alcoholic contents, glass types, and preparation guidelines are all included in the dataset. Based on the components, we created algorithms to categorize cocktails as either alcoholic or nonalcoholic, forecast their category, and suggest different kinds of glasses. The results give useful tools for customization in the beverage business, as well as information about cocktail trends.

1. Introduction

As a major component of the drink industry, cocktails which contain alcohol are distinguished by their variety of flavors, components, and presentation. Machine learning presents a chance to evaluate and forecast cocktails’ features of being alcohol or non-alcoholic using available data, which is in line with the growing interest in automation and customization. Conversely, non-alcoholic drinks, sometimes referred to as mocktails, have been becoming more popular due to a growing movement towards healthy living choices. Many customers are choosing non-alcoholic beverages as a result of a growing understanding of the harmful effects of alcohol on their health. Pregnant women, designated drivers, and people abstaining from alcohol for personal or health reasons can all enjoy these cocktails, since they provide the same elegant and tasty experience as their alcoholic equivalents without the alcohol. Because it appeals to the growing market of health-conscious consumers, the non-alcoholic cocktail business is predicted to grow more quickly than the alcoholic cocktail market. In order to complete classification tasks and obtain knowledge that can improve decision-making in this field, this study intends to use a structured dataset of cocktails which contain alcohol. The goal is to focus on cocktails in bar or sectors in which alcohol is used. The research also analyzes how harmful alcohol is, and discusses the pros and cons of the alcohol used.

2. Literature Review

One of the main causes of liver-related death worldwide is alcohol-related liver disease (ARD), which is connected to excessive drinking and impacted by lifestyle and genetic factors. Although there are few integrated clinical trials treating AUD and ARD, studies indicate that dropping alcohol use improves liver health [1]. As shown in Figure 1, the cocktail drinks follow the same scheme.

There are serious public health issues associated with excessive drinking. Prior studies showed that a 12-week non-alcoholic beverage intervention decreased heavy consumers’ alcohol intake. This study investigated how the effectiveness of the intervention varied by gender [3] using the Mojet model, which divides components into product, individual, and environmental categories, to investigate the factors influencing consumer behavior with reference to no- and low-alcohol (NOLO) drinks [4]. In Poland, non-alcoholic drinks are becoming more and more popular as a result of shifting consumption patterns and growing health consciousness. Reference [5] examines how non-alcoholic drinks (NABs) affect different patient populations. As healthier substitutes, these drinks, which resemble alcoholic ones but have less than 0.5% alcohol, are becoming more and more well-liked. However, little is known about their effects on health, which results in a range of suggestions from medical professionals. Reference [6] investigates the effects of acute alcohol consumption and working memory capacity (WMC) on auditory selective attention, particularly the “cocktail party effect”, i.e., the ability to focus on one speaker in a crowded environment while also noting salient inputs like hearing one’s name [7]. Through processes like oxidative stress, sympathetic activation, and baroreceptor stimulation, alcohol use is known to contribute to hypertension. The effects of moderate social drinking, especially in non-alcoholics, are less well studied than the well-established link between heavy drinking and blood pressure [8]. In [9], the authors examine 117 case examples from the advertising sector from the 1980s to 2016. It brings attention to the ways that food and beverage advertising uses techniques including emotional appeals, celebrity endorsements, and health claims to increase sales and brand loyalty. Children were the target of almost 40% of campaigns, which frequently promoted unhealthy foods that were rich in fat, sugar, or salt (HFSS). No/low drinks are promoted as harm reduction tools for reducing alcohol-related health issues. Studies suggest that these drinks are more popular among affluent, health-conscious individuals, aligning with trends in socially advantaged demographics [10]. Although health authorities advise abstinence during pregnancy, many women continue to drink before and during the early stages of pregnancy, which is connected with negative outcomes such as low biological weight and fetal alcohol syndrome [11]. Alcohol’s role in obesity is complex, with mixed evidence on whether it significantly contributes to weight gain. Public health policies, like nutritional labeling, aim to address the global obesity epidemic by promoting informed consumer decisions [12]. Alcohol consumption is firmly established in social and cultural behaviors but raises considerable public health issues due to its role in diseases, accidents, and societal expenses. The idea that moderate alcohol use has health advantages is being challenged by more and more research [13]. Young people’s harmful alcohol consumption is a serious public health concern, since it frequently results in mishaps, injuries, and alcohol-related illnesses. Web-based programs and motivational interviewing are examples of current therapies that do not work in real time during drinking episodes [14]. Both alcoholic and non-alcoholic drinks are associated with gendered stereotypes, such as the following:

• Masculine: Whiskey, beer, soda, energy drinks, and protein shakes.
• Feminine: Wine, mixed drinks, diet soda, iced tea, and fruit drinks.

3. Methodology

The dataset has about 5000 entries of drinks and is organized into 11 columns. Among the important columns are “id”, which is each cocktail’s unique identification name, and “The cocktail’s name. alcoholic”, which indicates if the cocktail contains alcohol or not. The cocktail’s category (such as cocktail, shot, or ordinary drink) is indicated by category: “glassType”.

The type of glass that the cocktail is served in, the ingredients, and their corresponding measurements are detailed in “ingredients” and “ingredient Measures”. This dataset improves our comprehension of cocktails’ qualities by enabling a variety of classification and prediction tasks [15,16,17].

The dataset is used in this study to carry out classification tasks using machine learning methods. One of the main goals is to categorize cocktails as either alcoholic or non-alcoholic according to their ingredients and category [18].

We determine a cocktail’s categorization based on its components and glass type and suggest glass types according to the components or category of the cocktail. Categorical columns were encoded, and missing values were handled for preprocessing [19]. For classification problems, models including, Decision Trees, Support Vector Machines, Random Forest, and Naïve Bayes were used.

4. Result

The results demonstrate machine learning’s promise in the beverage sector. The models can make suggestions that suit user preferences by examining the connections between components, glass kinds, and categories [20]. Managing partial or missing data and guaranteeing model generalizability across various datasets are significant challenges. The bar chart illustrates the accuracy performance of four different machine learning classification algorithms: Decision Tree, Support Vector Machine (SVM), Random Forest, and Naïve Bayes. Accuracy is a key metric that indicates the proportion of correctly predicted instances out of the total predictions made. The chart is designed to help compare how well each model performs in terms of predictive accuracy.

From

Table 1. Different models and their accuracies.

Model	Accuracy
Decision Tree	95%
Support Vector Machines	92%
Random Forest	88%
Naive Bayes	59.6%

we can see, the Decision Tree algorithm achieves the highest accuracy of of 95%. This indicates that it performs well on the given dataset, effectively learning patterns and making correct predictions. The Support Vector Machine and Random Forest models follow closely, both achieving around 90% accuracy. These results suggest that these models are also reliable for the classification task, with SVM being known for its ability to handle high dimensional data and Random Forest for its ensemble nature, which helps reduce overfitting and improves robustness.

On the other hand, the Naïve Bayes classifier shows a significantly lower accuracy, approximately 60%. This drop in performance, as shown in Figure 2, could be due to the assumptions made by the Naïve Bayes algorithm, mainly that features are conditionally independent. In many real-world datasets, this assumption does not hold true, which can limit the model’s effectiveness. While Naïve Bayes is fast and works well with small datasets or specific types of problems (like text classification), it may underperform compared to more complex models on structured data.

5. Conclusions

This study shows how machine learning may be used to analyze and predict cocktails which are alcoholic or non-alcoholic, as well as their properties or qualities. The findings open up new paths for investigation into inventory optimization and handmade processes in the beverage sector. Predictive accuracy could be further improved by investigating cutting-edge deep learning algorithms and adding more varied cocktails to the dataset.

This study is limited by the imbalance or lack of diversity of the data used for training the models. If the data contains more alcoholic cocktails compared to non-alcoholic cocktails (or vice versa), the model will be biased towards predicting the majority class, and the results will be skewed. Also, the data lacks more complex features such as flavor profiles, ingredient ratios, or cultural aspects, which might be extremely relevant for distinguishing more complex cocktail types. These exclusions can affect the model’s ability to learn more complex distinctions, rendering its predictions ultimately less accurate. A second concern is the application of simpler machine learning models like Decision Trees and Naïve Bayes, which are interpretable and easy to apply but might not be able to grasp deeper, non-linear relationships between features that other models, such as deep learning models (specifically convolutional ones) might.

Neural networks (if visual inputs are present in the data) or recurrent neural networks (for text descriptions) might express more intricate patterns hidden in cocktail recipes or ingredient lists.

For future work, the incorporation of multi-modal data—i.e., images, ingredient descriptions, preparation steps, or even customer reviews—can potentially enhance the performance and flexibility of prediction models. Transfer learning techniques and pre-trained models for similar domains can also give a head start in pushing classification tasks forward. Lastly, the incorporation of real-time analysis of user preferences and consumption habits can lead to dynamic and personalized recommendations for consumers and bartenders alike. Scaling the project to a recommendation engine that is integrated into hospitality or e-commerce sites would be a very valuable application. Overall, while the current study is a promising beginning, future work is hoped for in the areas of richer data, more complex models, and integration into real-world applications in the food and beverage industry.