Self-adaptive Parameters Optimization for Incremental Classification in Big Data Using Swarm Intelligence

Abstract

Nowadays, big data is one of the most technical challenges confront researchers and companies. The main challenge lies in the fact that big data sources usually formed in a continuous data stream. Thus, many previous researches present incremental data mining approaches to deal with the challenges of the data streams by adapting traditional machine learning algorithms. Artificial Neural Network (ANN) is a common technique in this field. The main challenge is how to optimize the neural network parameters to deal with a huge data arrive over time. These parameters, which are vital for the performance of a neural network, are called hyperparameters. Most earlier optimization approaches have dealt with big data containers instead of big data streams or handled big data streams with time consumed. This paper proposes an incremental learning process for ANN hyperparameters optimization over data stream by utilizing Grasshopper Algorithm (GOA) as a swarm intelligence technique. GOA is utilized to make a balance between exploration and exploitation to find the best set of ANN hyperparameters suitable for data stream. The experimental results illustrate that the proposed optimization model yields better accuracy results with appropriate CPU time.