Dynamic Induction Model for Students Behavior Analysis

Abstract

The volume of data is growing rapidly due to the usage of social sites like twitter, facebook etc. 80% of the college students spend their maximum time on social media. They share their views, feelings, emotions on it. This massive data is useful for institutes for getting feedback about any student or services provided by them. This feedback will help institutes to provide proper mentoring to students or to take any corrective action which will improve quality of service. The use of machine learning algorithms for analyzing this data will add more knowledge into the knowledge of institutes. Decision tree algorithm provides visual representation of data which is useful for social media data analysis. Traditional machine learning algorithms like C4.5 or CART have a limitation of memory size because they store all data on memory for building a model. So, these algorithms are not suitable for large volume of data. These algorithms performs best if the size of data is small but if size of data increases the same algorithms shows poor results. In this paper, we have used Hoeffding tree for large volume of data and proved with results that Hoeffding tree performs best against other Machine learning algorithms. Other algorithms like SVM, Naïve Bayes, Decision Tree C4.5 work well if the data set is small but their performance degrades if data size increases. To increase accuracy, we have used different classifiers at leaf level and analyzed different split criteria’s. We have collected dataset from twitter social site. Different phases of social media data mining are also explained in detail.

Keywords: Decision trees, Hoeffding trees, Social media data

View PDF