Toward a Monotonicity Constrained Support Vector Regression Model

Abstract

Machine learning techniques are widely used for analysis and extraction of knowledge Data mining as a tool of machine learning for knowledge discovery in databases (KDD) can automatically or semi-automatically analyze large quantities of data In recent years support vector machine (SVM) a state-of-the-art artificial neural network based on statistical learning has been the focus of research in machine learning due to its excellent ability Support vector regression (SVR) is the most common form of application of SVMs when the output is continuous Instead of minimizing the observed training error SVR attempts to minimize the generalization error bound so as to achieve generalized performance SVR has been applied in various fields – time series and financial (noisy and risky) prediction approximation of complex engineering analyses convex quadratic programming and choices of loss functions etc However in many real-world problems incorporating prior knowledge into SVR can improve the quality of models that are only data-driven and close the wide gap between academic and business goals We can observe some monotonic relationships between the output value and attributes and it has been shown that a technique incorporating monotonicity constraints can reduce errors In this study we propose a knowledge-oriented new support vector regression model with monotonicity constraints and exploit the knowledge of experts to retrieve monotonic rules from datasets After which we construct monotonicity constraints to implement the proposed regression model Experiments conducted on function prediction and real-world data sets show that the proposed method which is not only data driven but also domain knowledge oriented can help correct the loss of monotonicity in data during the collection process and performs better than traditional methods

Date of Award	2014 Jun 20
Original language	English
Supervisor	Sheng-Tun Li (Supervisor)