A comparative study of support vector machine and logistic regression article · January 021 citations reads 11 authors

Keywords: 
Machine learning, Support Vector Machine, Logistics Regression, classification, Regression. 
1.
Introduction 
The Support Vector Machine (SVM) and the logistic regression share common features, in that 
both statistical tools can be used for classification. However, as their names sound differently, their 
approaches to classification are different from each other. In other words, each of them has a 
unique and different approach to classification. For instance, while the SVM has a geometrical 
appeal because its concepts are based on hyper planes, the logistic regression has a strong 
connection to regression. It is therefore considered imperative that in this study, the different tools 
should be studied, with a view to uncovering their possible similarities and differences. For this 
reason, in the subsequent sections that follow, we shall look at both tools as examples of machine 
learning tools. We shall further discuss what classification means and clearly introduce the support 
vector machine and the logistic regression as the classification tools this study focuses on. 
1.1
Classification 
Classification is one example of supervised learning and another one is regression, but we are not 
concerned with regression in this study. The objective is to produce a classification function or 
classification model based on features of the data set to classify the information in the data properly 
and efficiently. The classification model can group unknown samples into a given category. It is 

---

O
RIENTAL JOURNAL OF SCIENCE & ENGINEERING VOL -2, ISS-1, FEB - 2021 
www.ojse.org
ojse©2019 
Page 86 
common knowledge that both classification and regression can be used for prediction. However, 
unlike regression methods, the output of the classification is a discrete value, while the output of 
the regression is a continuous or ordinal value. Many problems in real life can be converted into 
classification problems by using binary codes. In a binary classification problem, we try to predict 
whether the result belongs to one of two classes, such as true or false, yes or no, die or survive etc. 
For example, a classification model can be constructed by customer classification to perform risk 
evaluation on bank credit card business. Other classification applications exist which includes, 
image recognition technology and automatic text classification techniques in search engines (Xu, 
2018). 
Classification problems and methods have been considered a key part of machine learning, with a 
lot of applications published in recent years. The idea of classification in Machine Learning has 
been traditionally treated in a broad sense, very often including supervised, unsupervised, and 
semi-supervised learning problems. Unsupervised learning is allowed to find out hidden patterns 
and information in the data sets that wasn’t visible. This paradigm is especially useful to analyze 
whether there are differentiable groups or clusters present in the factor (e.g., for segmentation). In 
the case of supervised learning, however, each data input object is reassigned a class label. The 
main task of supervised algorithms is to learn model that ideally produces the same labelling for 
the provided data and generalizes well on unseen data (i.e., prediction). In this regard, classification 
learning applications are widely used to cope with difficult problems arising from real-life 
activities (Wang and Zhen, 2014). 
When the researchers are faced with the classification of the data set, they usually apply their 
desired model. This is determined by their knowledge of the available models. Different 
classification models might produce different classification results, and the quality of the models 
will proportionally affect the accuracy of the classification results and the organization of machine 
learning tool used. Therefore, when classifying big data, it is crucial to choose the most appropriate 
classification algorithm (Xu, 2018).