An improved active learning sparse least squares support vector machines for regression

Recently, active learning sparse least squares support vector machines (AL-LSSVM) was put forward to solving the sparseness problem of least squares support vector machine, which demonstrates better sparseness and robustness than Sukens´ algorithm in removing the similar samples and solving the problem of heteroscedasticity. However, there are several problems to consider. In choosing support vectors, the approximation problem was solved by recursively selecting data with a big error in the previous process. And the performance will be not predicted to the best after the data has been selected. As a result, the process of selecting is longer and instability, its performance is sensitive to the choosing data. Therefore, an improved active learning least squares support vector machines (IAL-LSSVM) is introduced, with selecting samples based on sorted performance spectrum, which gradually chooses the support sample with the best performance after select the sample for the next iteration. In order to prove the efficacy and feasibility of our proposed IAL-LSSVM, some experiments are done comparing to AL-LSSVM. And they are all favorable for our viewpoints. That is, the IAL-LSSVM has better sparseness and robustness than AL-LSSVM.