A Bi-Level Approach for Hyper-Parameter Tuning of an Evolutionary Extreme Learning Machine

One of the critical challenges in the implementation of machine learning algorithm is hyperparameter optimization as performance of any machine learning model is sensitive to the setting of their hyperparametersr. Evolutionary Algorithms (EA) is widely used for hyperparameter optimization due to its efficient intellectual tuning strategies. The time complexity is appreciably changed with respect to the size of dataset used for training. On the other hand, large dataset is required for pursuing the better prediction. In this paper, we have proposed a methodology namely Bi-Level Evolutionary Extreme Learning Machine(bL-EELM) based on bi-level programming approach for tuning hyperparameter of an Evolutionary Extreme Learning Machine(EELM). we divided our problem into two levels. We consider an E-ELM module as a lower level optimization problem. In our upper level we placed a evolutionary module whose task is to create a population of hyperparameters and feed to lower Level as an input of EELM. We have chosen ten benchmark classification problems for the experiment and analysis of our proposed approach. Experimental results proofs that our proposed approach has better prediction accuracy as well as generalization performances compare to Extreme learning machine(ELM) and EELM.