BİLDİRİ DETAY

Hasan YILDIRIM, M.Revan ÖZKALE
A NEW EXTREME LEARNING MACHINE ALGORITHM USING LIU REGRESSION
 
Extreme learning machine (ELM) algorithm is proposed as a new algorithm for single layer feed forward neural networks and attracted attention because of computational easiness and satisfying generalization performance. On the other hand, these improved properties are weak in the presence of perturbation, i.e multicollinearity. A novel ELM algorithm based on Liu regression estimator (L-ELM) which is effective for handling issues related with multicollinearity is proposed in this study. Furthermore, different selection methods for Liu biasing parameter are introduced. The proposed algorithm is compared with the preliminary ELM using a variety of benchmark datasets. According to the experimental results, it is obtained that L-ELM for at least one Liu biasing parameter improves the generalization performance of ELM, and also outperforms ELM in terms of stability performance. Furthermore, the percentage of improvement on the basic ELM algorithm has been computed for each type of Liu biasing parameter and dataset. It has been seen that the selection method of Liu biasing parameter is data dependent and dramatically effective on the performance of L-ELM. Consequently, even if the proposed algorithm needs a little more time than ELM on training, it can be used as a stable and efficient alternative tool to ELM in regression studies.

Anahtar Kelimeler: Extreme Learning Machine, Machine Learning, Neural Networks, Liu Estimator Regression



 


Keywords: