Stream Mining for Network Sensor Data Using Classification and Metaheuristic Techniques

Abstract

Datastream mining comes to play an integral role in real-time applications. The main sources of the data stream flows are sensors, multimedia and social media. It contains remote sensors, stock markets, tweets, and video surveillance systems. The data stream has distinctive characteristics which include its high speed and very huge volume, in addition to its ability to change over time. It has many challenges, one of which is the concept drift that happens due to the continuous property of data stream. Traditional data mining techniques could not deal with or mine this big, rapid data. The aim of this work is to classify sensor data using metaheuristic and classification techniques. The metaheuristic approach is used to build the balanced chunks of data using Particle Swarm Optimization (PSO) to obtain a better classification accuracy. The meta model formed by the metaheuristic and classification techniques shows an improvement in accuracy performance at a high rate, as compared to non-metaheuristic models. Multiple classifiers have been chosen based on fuzzy function for each chunk to select the optimal one which gives the optimal results with high accuracy. The obtained results show a good performance in terms of classification accuracy for the neural network with 90% and low positive rate.