An Efficient Feature Selection Scheme for Big Data Based on GOA and LSTM

Abstract

The traditional feature selection models are assumed with all the instances of big data and features are used for learning. Therefore, in the case of real-life applications, it was not likely to face the data streams with feature streams or both. The present streaming feature selection algorithms focused mainly on eliminating redundant and irrelevant features. However, while selecting the most relevant features, the interacting features were ignored in the existing approaches. These features are essential for the proposed work that uses the Grasshopper Optimization Algorithm (GOA) for the relevant feature selection and Long Short Term Memory (LSTM) for the classification of diseases. The proposed work uses 4 types of datasets Leukemia, Colon, Prostate, and Breast Cancer. The GOA optimization problems in various domains find input parameters or the arguments for functioning. The minimum and maximum output functions have been obtained from the LSTM model which provides a large range of parameters like input biases, output biases, learning rate, and complexity. The results show that the Leukemia dataset which is 99.17% of accuracy with a execution time of 0.5472s better when compared to the existing streaming feature selection method obtains 82.26% of accuracy with a execution time of 2.4077s.