Research Article | Volume 5 Issue 2 (April-June, 2024) | Pages 1 - 10
Classification and Evaluation of the COVID-19 Dataset Based on Supervised Machine Learning Techniques
1
College of Biomedical Informatics, University of Information Technology and Communications, Baghdad, Iraq, 10011
Under a Creative Commons license
Open Access
Received
Feb. 15, 2024
Revised
Feb. 25, 2024
Accepted
March 28, 2024
Published
April 30, 2024
Abstract

To predict future events through supervised machine learning using algorithms. All of this requires smart systems through which the algorithms can be implemented. In this research paper, we used several classification methods for supervised machine learning (ML). An analysis of the algorithms was conducted and they were compared and evaluated. The best algorithm in the classification process was identified through the characteristics of the data set and variables. An evaluation of several machine learning algorithms was conducted in the WEKA application: Naïve Bayes (NB), J48, Linear Regression (LR), and K-Nearest Neighbors (KNN) as well as SVM, by using data for the emerging coronavirus (COVID-19). The results showed that SVM is better than other algorithms used in terms of accuracy. The results extracted from the algorithm have been proven to be effective in prediction, therefore supervised machine learning algorithms must be accurate, precise, and error-free.

Keywords
Machine Learning Techniques
Classifiers
Data Mining methods
SVM
Introduction

In essence, data mining is turning unprocessed observations into insightful knowledge. With the use of theory or hypothesis analysis, this data can be used to make more predictions. The scientific technique that underpins data mining—from which global insights are derived—has been successfully applied since the Renaissance in the West and even earlier in other countries. But the exponential increase in processing power in the last few decades, along with the widespread use of digital data, has fundamentally altered the terrain. The abundance of data that is


Now available has not only improved our comprehension of the natural world but also spawned a digital world where finding fresh insights in old data is a major endeavor. Our work revolves around this procedure, which is called knowledge discovery in databases (KDD) and is also known as data mining [1-3].

 

The field of computer science known as machine learning has emerged as one of the most quickly evolving subfields, and its ramifications are particularly far-reaching. In this context, it refers to the process of automatically recognizing meaningful patterns within data. The goal of technologies that are based on machine learning is to make it possible for programs to learn and get better [4]. Due to the fact that machine learning has become a mainstay of information technology, it has become a relatively big component of our lives, despite the fact that it is frequently buried. The growing volume of data suggests that intelligent data analysis will become more important in technology development.

 

One of the most significant applications of machine learning (ML) is data mining, which is among the many applications that include ML. On a regular basis, individuals tend to commit errors when conducting research or, perhaps, when seeking to establish correlations between a large number of characteristics [4-5]. Data mining and ML processes are similar to Siamese twins, and effective learning algorithms can produce a wide range of information. Data mining and machine learning have made significant strides in recent years as a direct result of the introduction of smart and nanotechnology. This has led to an increased interest in uncovering hidden patterns within data in order to extract beneficial information.

 

The combination of areas such as statistics, machine learning, information theory, and computer science has led to the development of a robust field that is characterized by a robust mathematical foundation and incredibly powerful tools. In the field of machine learning, algorithms are categorized into several groups according to the outcomes that are expected from the algorithm. A function that is capable of converting inputs into outcomes that are desired is developed through supervised learning. As a result of the extraordinary amount of data that has been collected over time, machine learning algorithms have developed throughout time. Consequently, this has led to the implementation of a wide range of techniques for both supervised and unsupervised machine learning. On account of the fact that the objective is typically to teach the computer to learn a classification system that we have built, supervised learning is a technique that is utilized quite commonly in classification difficulties [6].

 

The classification problem is a popular construction of the monitored knowledge task. In this task, the learner is required to learn (or estimate the performance of) an equation that maps a vector into one of multiple groups by analyzing a large number of input-output samples of the function. When it comes to ML, inductive computational learning refers to the procedure of obtaining a set of rules from cases, which are instances that are included in a training set. In a broader sense, it refers to the development of an algorithm that can extrapolate from fresh occurrences. The manner in which supervised ML is utilized to find a solution to a real-world problem is illustrated in Figure 1.

 

 

Figure 1: The General Supervised ML Procedures [10, 17]

 

This endeavor tries to categorize machine learning algorithms and determine the most effective and efficient way that possesses high accuracy and precision. This study compares the recital of several procedures on large and small datasets, with the goal of correctly classifying them and providing insights into building supervised machine learning models.

 

The following sections of this work are organized as follows: Section 2 provides a literature overview on classification of supervised learning algorithms, followed by a methods section, results section, and conclusion section with recommendations for further research.

 

 

Related Works

  1. Taxonomy of Supervised Learning techniques

As stated in [7-9], supervised machine learning techniques that focus more on classification involve a number of them: Decision Tree, Random Forest (RF), Bayesian networks, and so forth; naïve bayes classifier, logistic regression, support vector machine; and so forth.

 

  • For the purpose of classifying input vectors, a linear classifier, which is a type of linear model for classification, employs linear (hyperplane) decision boundaries [11]. When it comes to machine learning, the primary objective of classification in linear classifiers is to differentiate between various categories by categorizing items that have similar feature values. A linear classifier, according to [11], accomplishes this goal by classifying data according to the linear combination of characteristics' values. Given their reputation for quick processing, linear classifiers are recommended in situations where classification speed is crucial. They also perform well in situations when the dimensionality of the data is high, like in text classification, where each element regularly corresponds to the word counts in a document. However, the tolerance determines dataset variable convergence speed. The margin of error indicates a dataset's linear separation or problem-solving tractability.

     

  • Logistic regression is a categorization technique that services a single polynomial logistic regression with a single estimator and builds its model using classes. Typically, logistic regression indicates the location of the class boundaries and the dependence of probability distributions on distance starting from the border, in a certain direction. A larger data set causes this to drift towards the immoderations (0 and 1) more quickly. These probabilistic claims elevate logistic regression above the status of a simple classifier. It can be fitted in a different method and produces stronger, more accurate predictions, but those exact predictions could be off. A prediction method similar to Ordinary Least Squares (OLS) regression is logistic regression. On the other hand, prediction yields a dichotomous result when using logistic regression [12]. One of the most popular methods for the discrete analysis of information and utilized statistics is the logistic regression approach. Linear interpolation is used in logistic regression [12].

 

  • Support Vector Machines, sometimes known as SVMs, are the most cutting-edge approach to supervised machine learning [13]. SVM models and the classic perceptron-based multilayer neural networks have a lot in common because of their similarities. In support vector machine (SVM) theory, the concept of a "margin"—that is, either side of a hyperplane that divides two data classes—is crucial. By generating the maximum distance between the hyperplane that divides it and each example on the other side of it, it has been proved that maximizing the margin reduces an upper bound on the predicted generalization error [14]. This is accomplished by generating the greatest distance among the examples.

     

  • Decision Trees: Decision Trees (DT) are branches that use feature values to rank instances to classify them. In a decision tree, each node indicates a feature in an instance that needs to be classified, and each branch indicates a possible value that the node could take. The trees are organized in a hierarchical structure. The decision tree intends to classify the features. Assuming that the root node is the starting point, examples are categorized and sorted in accordance with the feature values of each other instance [9]. In the field of decision tree learning, which is utilized in the fields of data mining and ML, a decision tree is utilized as a prediction model. Through this process, explanations about an item are mapped to judgments about the goal value of the item. The phrases "regression trees" and "categorization trees" are more appropriate to use when referring to structures of this kind. DT often uses post-pruning approaches, which assess decision tree performance as it is cropped using a justification set. Any node can be eliminated and given the training example category that is ordered to it that is most frequently encountered [9, 14].

 

  • Naive Bayes (NB) classifiers are gaining recognition as crucial instruments in the domains of machine learning and predictive modeling. The present study examines the practical applications and theoretical foundations of Naive Bayes algorithms across numerous fields of study. For classification problems, the Naive Bayes method, which is founded on the principle of Bayes and the notion of features becoming independent, is straightforward and effective [15-16].

     

  • Supervised AI application K-Nearest Neighbors (KNN) is used to address classification and regression challenges. It is based on the principle of nearestness, which works to label an unobserved data point using its nearest neighbors, that is, it


 

uses an overwhelming class, such as data points located in spaces that are distinct and also of interest [18].

  • Logistic Regression (LR), is a binary classification of ability in supervised learning. It is considered a linear model. Logistic regression predicts binary outcomes by using a set of variables that are expected. This comes because of its ability to analyze and classify. The probability scores are between 0 and 1 through the use of the logistic function. The algorithm has predictive capabilities [19].

 

  1. Characteristics of Machine Learning Algorithms

 

Supervised machine learning methods are used in a variety of fields, as indicated by several ML application-oriented articles [20] and [6]. SVMs and neural networks outperform other algorithms when dealing with multidimensional and incessant topographies. On the other hand, logic-based systems achieve more effectively when employing discrete or categorical characteristics. NB may be able to function with very few datasets, in contrast to neural network models and SVMs, which require huge sample numbers to attain maximum prediction accuracy.

 

There is a consensus among many people that the k-Nearest Neighbors (k-NN) algorithm is particularly sensitive to irrelevant characteristics, which is a property that is inherent in the way it operates. In addition, the presence of elements that are not relevant might dramatically diminish the effectiveness of neural network training, rendering it impractical in some situations. Because they tend to generate hyper rectangular sections that are orthogonal to one variable axis and orthogonal to others, the majority of decision tree algorithms have difficulty dealing with problems that need diagonal division. In contrast, SVMs perform well in scenarios with multicollinearity and nonlinear correlations between input and output features. Understanding these distinctions is critical when choosing the best machine-learning algorithm for a specific problem area.

 

NB utilizes little storage capacity during learning and grouping, with the strict minimum being the memory needed to record prior and future provisional prospects. The KNN technique requires a significant amount of storage space for training and an equally large execution space. Non-lazy learners typically have a smaller execution space than training space, as their classifier is a simplified presentation of data. Naive Bayes and KNN are more suited for incremental learning than rule-based algorithms.

 

DT and NB have diverse operative profiles, with one exhibiting excellent accuracy while the other may not. In contrast, DT and rule analyzers have comparable operational features. Similarly, SVM exhibits similar functional tendencies. It's vital to remember that no learning machine can continuously beat others across all datasets. The performance of algorithms is determined by variables such as dataset properties, variable kinds, and the number of instances. The No Free Lunch hypothesis states that no particular learning approach may surpass others on all datasets.

Methodology

In accordance with our results in investigation, a total of COVID were utilized, and the results of the tests were discovered to be both positive and negative. A wide range of information concerning the results of this experiment may be found in the COVID data, which can be accessed at the following website: https://ourworldindata.org/coronavirus. Through the utilization of WEKA, a comparative examination of four distinct supervised machine learning algorithms was carried out for the purpose of. The dataset was built with a single nominal attribute column serving as the principal component. The dependent variable was the one that was


 

being constructed. At the same time values of 1 in the class circulation were replaced with "YES," which is a representation of the fact that the patient was evaluated positively for covid, values of 0 were replaced with "NO," which indicates that the patient was evaluated adversely for covid. This update was necessary since the majority of algorithms require that each column contain at least one nominal variable. This made it vital to make this modification. During the research, a variety of classification techniques were utilized, including Decision Table, RF, NB, SVM, KNN, LR, and J48 in particular. During the course of the comparative analysis, a variety of characteristics were taken into consideration. These characteristics included things like the time, the appropriate classification, the accidental classification, the test mode, the number of examples, and so on.

 

The accuracy and precision of several machine learning approaches were evaluated in this study through the use of parameter adjustment, which was applied to the sets of examples.

Outcomes and Discussions

A. Outcomes.

WEKA was used to classify and compare several ML methods.

 

Table 1: Alternative Algorithms for Accuracy Performance

 

Methods

Accuracy

SVM

83.21

RF

77.11

NB

82.01

J48

57.10

KNN

52.21

LR

66.99

 

Table 2: Precision of Utilizing Datasets and Various Approaches

Methods

Precision

SVM

0.89

RF

0.77

NB

0.79

J48

0.69

KNN

0.62

LR

0.78

 

Figure 2: Assessment of Accuracy for Datasets

 

Figure 3: Assessment of Precision for Datasets

Discussion

The support vector machine (SVM) algorithm is stable and effective in a variety of machine learning contexts by the fact that its strong accuracy performance maintains consistency across datasets of diverse sizes, as shown in Tables 1 and 2 and Figures 2 and 3. It has also been proven in several different machine-learning situations. Proving the accuracy and robustness of an algorithm using data sets of different sizes and complexity. SVMs are powerful algorithms that can be applied to a wide range of modeling, classification, and forecasting problems. That is, it is reliable. Therefore, in addition to having the ability it can be applied in real-life situations. It is considered the most reliable in the machine learning arsenal.

Conclusion and Further Works

A huge amount of data instances and the fine-tuning of variables are both required for machine learning categorization. When it comes to developing a model for a computer algorithm, time is not the only factor to consider; accuracy and classification are also important. Utilizing the most effective learning algorithm for a certain data set does not guarantee the same level of precision and accuracy for another data set that possesses distinct characteristics. When it comes to ML arrangement, the most important question is not whether one algorithm is superior to others; rather, it is the circumstances under which a particular approach might meaningfully outperform others on a particular application challenge. The learning aims to identify functions that correlate datasets with the efficiency of algorithms. The learning identifies relationships among meta-attributes, which indicate task features, and learning algorithm efficacy. Learning tasks have various properties, such as the number of cases, fraction of categorical properties, missing values, and class volatility. This study suggests using shared processing for massive data sets. This allows for significant associations between variables, leading to improved model outputs.

References
  1. Mohd Selamat, Siti Aishah, Simant Prakoonwit, and Wajid Khan. "A review of data mining in knowledge management: applications/findings for transportation of small and medium enterprises." SN Applied Sciences 2.5 (2020): 818. https://link.springer.com/article/10.1007/s42452-020-2589-3 
  2. Palacios, Carlos A., et al. "Knowledge discovery for higher education student retention based on data mining: Machine learning algorithms and case study in Chile." Entropy 23.4 (2021): 485. https://www.mdpi.com/1099-4300/23/4/485 
  3. Maimon, Oded, and Lior Rokach. "Introduction to knowledge discovery in databases." Data mining and knowledge discovery handbook. Boston, MA: Springer US, 2005. 1-17. https://link.springer.com/content/pdf/10.1007/0-387-25465-X_1.pdf 
  4. Shafiq, Muhammad, et al. "Data mining and machine learning methods for sustainable smart cities traffic classification: A survey." Sustainable Cities and Society 60 (2020): 102177. https://www.sciencedirect.com/science/article/pii/S2210670720301645 
  5. Ananey-Obiri, Daniel, and Enoch Sarku. "Predicting the presence of heart diseases using comparative data mining and machine learning algorithms." International Journal of Computer Applications 176.11 (2020): 17-21. https://www.researchgate.net/profile/Daniel-Ananey-Obiri/publication/340678267_Predicting_the_Presence_of_Heart_Diseases_using_Comparative_Data_Mining_and_Machine_Learning_Algorithms/links/5e98702092851c2f52a86a4a/Predicting-the-Presence-of-Heart-Diseases-using-Comparative-Data-Mining-and-Machine-Learning-Algorithms.pdf 
  6. Yang, Xin-She. Introduction to algorithms for data mining and machine learning. Academic press, 2019. https://books.google.com/books?hl=en&lr=&id=SNKdDwAAQBAJ&oi=fnd&pg=PP1&dq=6-%09X.-S.+Yang,+Introduction+to+algorithms+for+data+mining+and+machine+learning.+Academic+press,+2019.&ots=i3yjJGoEMm&sig=UuGZpcDIPvfDSvJu-ldSlrh0B3Q 
  7. Palmer, Alfonso, Rafael Jiménez, and Elena Gervilla. "Data mining: Machine learning and statistical techniques." Knowledge-Oriented Applications in Data Mining, Prof. Kimito Funatsu (Ed.) (2011): 373-396. https://books.google.com/books?hl=en&lr=&id=CLqZDwAAQBAJ&oi=fnd&pg=PA373&dq=7-%09A.+Palmer,+R.+Jim%C3%A9nez,+and+E.+Gervilla,+%22Data+mining:+Machine+learning+and+statistical+techniques,%22+Knowledge-Oriented+Applications+in+Data+Mining,+Prof.+Kimito+Funatsu+(Ed.),+pp.+373-+396,+2011.&ots=9lCdtHz0fS&sig=zrzdLsLJ2InXS_4r7b-kl8rwacQ 
  8. Bonaccorso, G. "Machine learning algorithms Packt Publishing Ltd." (2017).
  9. Marsland, Stephen. Machine learning: an algorithmic perspective. Chapman and Hall/CRC, 2011. https://www.taylorfrancis.com/books/mono/10.1201/9781420067194/machine-learning-stephen-marsland 
  10. Lavecchia, Antonio. "Machine-learning approaches in drug discovery: methods and applications." Drug discovery today 20.3 (2015): 318-331. https://www.sciencedirect.com/science/article/pii/S1359644614004176 
  11. Ballarini, Dario, et al. "Polaritonic neuromorphic computing outperforms linear classifiers." Nano Letters 20.5 (2020): 3506-3512. https://pubs.acs.org/doi/abs/10.1021/acs.nanolett.0c00435 
  12. Connelly, L. "Logistic Regression." Medsurg Nursing, vol. 29, no. 5, (2020), pp. 353-354.
  13. Campbell, Colin, and Yiming Ying. Learning with support vector machines. Springer Nature, 2022. https://books.google.com/books?hl=en&lr=&id=h4RyEAAAQBAJ&oi=fnd&pg=PP1&dq=13-%09C.+Campbell+and+Y.+Ying,+Learning+with+support+vector+machines.+Springer+Nature,+2022.&ots=ZsUGF6i3IN&sig=Bsr4VtbVwgtzgT2fEMDTruZOQLQ 
  14. Sethy, Prabira Kumar, et al. "Deep feature based rice leaf disease identification using support vector machine." Computers and Electronics in Agriculture 175 (2020): 105527.https://doi.org/10.1016/j.compag.2020.105527 
  15. Berrar, Daniel. "Bayes' Theorem and Naive Bayes Classifier." (2019): 403-412. https://www.researchgate.net/profile/Daniel-Berrar/publication/324933572_Bayes'_Theorem_and_Naive_Bayes_Classifier/links/5d837aba92851ceb79143b04/Bayes-Theorem-and-Naive-Bayes-Classifier.pdf 
  16. Rish, Irina. "An empirical study of the naive Bayes classifier." IJCAI 2001 workshop on empirical methods in artificial intelligence. Vol. 3. No. 22. 2001. https://citeseerx.ist.psu.edu/document?repid=rep1&type=pdf&doi=2825733f97124013e8841b3f8a0f5bd4ee4af88a 
  17. Nasteski, Vladimir. "An overview of the supervised machine learning methods." Horizons. b 4.51-62 (2017): 56. https://www.researchgate.net/profile/Vladimir-Nasteski/publication/328146111_An_overview_of_the_supervised_machine_learning_methods/links/5c1025194585157ac1bba147/An-overview-of-the-supervised-machine-learning-methods.pdf 
  18. Cunningham, Padraig, and Sarah Jane Delany. "K-nearest neighbour classifiers-a tutorial." ACM computing surveys (CSUR) 54.6 (2021): 1-25. https://dl.acm.org/doi/abs/10.1145/3459665 
  19. Jiang, Lai, and Runming Yao. "Modelling personal thermal sensations using C-Support Vector Classification (C-SVC) algorithm." Building and Environment 99 (2016): 98-106. https://www.sciencedirect.com/science/article/pii/S0360132316300221
Recommended Articles
Research Article
Computer Driven Library Management and Service Rendering System: Mobile Library Landscape
...
Download PDF
Research Article
Basics of Foam Science – A Brief Review
Download PDF
Research Article
Ann.Si. EssSubmodules And Ann.Si.Max Submodules
Download PDF
Research Article
Use of Fractal Batik Software for Batik Motif Design in Semarang Small and Medium Enterprises
Download PDF
Flowbite Logo
PO Box 101, Nakuru
Kenya.
Email: office@iarconsortium.org

Editorial Office:
J.L Bhavan, Near Radison Blu Hotel,
Jalukbari, Guwahati-India
Useful Links
Order Hard Copy
Privacy policy
Terms and Conditions
Refund Policy
Others
About Us
Contact Us
Online Payments
Join as Editor
Join as Reviewer
Subscribe to our Newsletter
Follow us
MOST SEARCHED KEYWORDS
Copyright © iARCON Internaltional LLP . All Rights Reserved.