A Comparison between Linear and Non-Linear Machine Learning Classifiers

Abstract

Data mining can be defined as searching for similarities and patterns in a huge amount of data in a certain knowledge field and to arrange them in classes and clusters. Many classification algorithms and clustering techniques are implemented to suit different types of data such as numeric, real, and nominal data types. Each classification and clustering algorithms are implemented in a certain approach. Some are linear and some are non-linear algorithms. In this paper, a comparison between some linear and non-linear classification algorithms has been conducted to study the performance of these classifiers with three different types of data set. The first data set is the collected MRI images of the brain tumor with type real, the second is diabetes data set with type numeric and the third is the breast cancer data set with type nominal. The linear classifiers chosen for this study are Lazy and Bayesian classifiers. While for the non-linear both Multilayer Perceptron (MLP) and Linear Vector Quantization (LVQ) are chosen. The results showed that the performance of the nonlinear classifiers was better than the linear classifiers with all data sets. In particular the accuracy rate of both MLP and LVQ with the real brain tumor data set is 91%, 83% respectively. On the other side, the linear classifiers showed comparable result with all datasets.