Filter-Wrapper Feature Selection with Improved Deep Learning for Education Data Mining

Abstract

The quality of learning models has a huge impact in deriving useful information from data. For this scenario, preprocessing handles inconsistencies in data to improve the model’s accuracy. The foremost major steps in preprocessing are data Quality Assessment, Cleaning, Transformation, and Feature Selection prior to Machine Learning Algorithms. The first phase of this research work comprise of Feature Selection to have relevant features that gives the information about significant features also gains the accuracy and Supervised Classification in order to categorize the learners based on their performance. This work proposes Filter-Wrapper Feature Selection method. The method starts with assigning weights using Filter method to each Feature based on Symmetrical Uncertainty and creates a subset of Features with weights. Instead of applying all attributes to the Wrapper feature subset method, the attributes with higher weights which are obtain from Filter method are only applied in order to minimize the computation complexity and to improve accuracy. Generally, the wrapper Feature method includes a learning algorithm to have better feature as well as classifies the dataset. For the learning algorithm, this paper applies Deep Multi FeedForward Neural Network with AdaDelta Optimizer, neural network parameters. The performance is assessed in Rapidminer tool using the Classification evaluation measures namely Accuracy, Weighted Mean Recall (WMR), and Weighted Mean Precision(WMP).