A meta-heuristic based hybrid predictive model for sensor network data

Abstract

Many prediction algorithms and techniques are used in data mining to predict the outcome of the response variable with respect to the values of input variables. However from literature, it is confirmed that a hybrid approach is always better in performance than a single algorithm. This is because the hybridization leads to combine all the advantages of the individual approaches, leading to the production of more effective and much improved results. Thus, making the model a productive one, which is far better than model proposed using individual techniques or algorithms. The purpose behind this chapter is to provide information to the users on how to build and investigate a hybrid Feed-forward Neural Network (FNN) using nature inspired meta heuristic algorithms such as the Gravitational Search Algorithm (GSA), Binary Bat Algorithm (BBAT), and hybrid BBATGSA algorithm for the prediction of sensor network data. Here, FNN is trained using a hybrid BBATGSA algorithm for predicting temperature data in sensor network. The data is collected using 54 sensors in a controlled environment of Intel Berkeley Research lab. The developed predictive model is evaluated by comparing it with existing two meta heuristic models such as FNNGSA and FNNBBAT. Each model is tested with three different V-shaped transfer functions. The experimental results and comparative study reveal that the developed FNNBBATGSA shows best performance in terms of accuracy. The FNNBBATGSA under three different V-shaped transfer functions produced an accuracy of 91.1, 98.5, and 91.2%.