Multiple nonlinear integral for classification

Abstract

In this study, a new classification model - Multiple Nonlinear Integral with Multiple projections is proposed, which includes Double Nonlinear Integral extending to new variants for adapting for complicated data distribution and enhancing classification accuracy. When the performance is not satisfying by projecting with classical Nonlinear Integral, the second projection is need to stretch data in one dimension space to two dimension space. The value of two projection forms the 2-dimensional coordinates. All data in two-dimensional space can be classified by a straight line easily. The rest may be deduced by analogy, if the result is still not good for decision, the Multiple Nonlinear Integral can repeat n times double projections, in which n will be an optimized value to balance the performance and the complexity. The repeating can help adjust the data distribution in 2-dimensional space until being classified easily. The classification model based on Multiple Nonlinear Integral is applied to two kinds of datasets. One kind comes from the classical database; another kind is the real data about the HBV (Hepatitis B Virus) collected from hospital. The experimental results show that the new model has better performance compared with the classical algorithm and the classical Nonlinear Integral. Especially to the HBV data, Multiple Nonlinear Integral presents the superior on diagnosis to the others.