RNN BASED FEATURE SELECTION MODEL WITH NOISY FEATURE REMOVAL

The evolution of emerging technologies leads into the unbeatable data growth and this circumstance brings out the necessity of reduction in data dimensionality. Feature Selection is one of the dimensionality reduction techniques for identifying the most important features for learning is the base to machine learning. Real-world data, which is given as an input to the Machine Learning algorithms, are affected by many components; the presence of noise is a key aspect among them. The proposed Feature Selection Model incorporates Machine Learning with Recurrent Neural Network and Cloud Databases to achieve better Classification accuracy by identifying and removing noisy features in the Cloud environment. On the light of literature survey, most of the conventional Feature Selection Methods fail to remove noisy data exists in the feature. Eventually, the dataset retrieved from the Cloud Database may contain noisy data which may reduce the performance of classification. Hence, a new Feature Selection model “RNN based Feature Selection Model with Noisy Feature Removal” is proposed which removes noisy data and retains relevant, correlated features in Cloud environment. The proposed model utilizes the Kohonen Self-Organizing Map (KSOM) to identify and remove noisy features from the dataset retrieved from the Cloud Database. With the help of recursive path provided in the RNN approaches which is applied in the proposed model makes the Feature Selection process to be finely tuned in terms of accuracy by removing noisy features. The primary goal of this proposed model is to eliminate noisy features from the existing dataset and achieve Feature selection using RNN. The performance of the proposed Model is evaluated using Five Medical Databases retrieved from AWS Cloud. The experimental and evaluation results obtained on these Cloud databases show that a proposed Model eliminates noisy features with significant improvement in classification accuracy better than conventional Feature Selection Models.