A precise and stable machine learning algorithm: eigenvalue classification (EigenClass)
- PDF / 559,442 Bytes
- 12 Pages / 595.276 x 790.866 pts Page_size
- 8 Downloads / 213 Views
(0123456789().,-volV)(0123456789(). ,- volV)
ORIGINAL ARTICLE
A precise and stable machine learning algorithm: eigenvalue classification (EigenClass) Ug˘ur Erkan1 Received: 25 October 2019 / Accepted: 3 September 2020 Springer-Verlag London Ltd., part of Springer Nature 2020
Abstract In this study, a precise and efficient eigenvalue-based machine learning algorithm, particularly denoted as Eigenvalue Classification (EigenClass) algorithm, has been presented to deal with classification problems. The EigenClass algorithm is constructed by exploiting an eigenvalue-based proximity evaluation. To appreciate the classification performance of EigenClass, it is compared with the well-known algorithms, such as k-nearest neighbours, fuzzy k-nearest neighbours, random forest (RF) and multi-support vector machines. Number of 20 different datasets with various attributes and classes are used for the comparison. Every algorithm is trained and tested for 30 runs through 5-fold cross-validation. The results are then compared among each other in terms of the most used measures, such as accuracy, precision, recall, micro-Fmeasure, and macro-F-measure. It is demonstrated that EigenClass exhibits the best classification performance for 15 datasets in terms of every metric and, in a pairwise comparison, outperforms the other algorithms for at least 16 datasets in consideration of each metric. Moreover, the algorithms are also compared through statistical analysis and computational complexity. Therefore, the achieved results show that EigenClass is a precise and stable algorithm as well as the most successful algorithm considering the overall classification performances. Keywords Data classification Eigenvalues Learning algorithm Machine learning Supervised learning
1 Introduction The field of data classification is of a growing importance due to the unpredictability, large amount, and complexity of real-world data which include multi-class predictions in practical applications [1–3]. The evolution of a new classification algorithm is an essential and challenging research topic in the field of machine learning [4–6]. Classification methods aim to predict a class label of input samples include a set of attributes [7]. Classification methods determine class labels of observed input test data according to training data. In classifying data, various mathematical distance calculations and intuitive methods are employed, and expert opinions are considered [8]. Classification problems with multiple classes and nonlinear class & Ug˘ur Erkan [email protected] 1
Department of Computer Engineering, Engineering Faculty, Karamanog˘lu Mehmetbey University, 70200 Karaman, Turkey
constraints with considerable numbers of training data which lead to computational cost generally require complex classifiers [9]. In classification methods, a class is assigned to an observed input test data by performing a learning process with training data. The learning process is divided into two categories, i.e. supervised and unsupervised learning [10
Data Loading...
