Exploring Feature Selection Technique in Detecting Sybil Accounts in a Social Network

EasyChair Preprint 2240

Abstract

The amount of data is increasing rapidly. With the advent in technology social networks are becoming popular day by day. Machine learning provides us the methods to extract useful information. There are different machines learning techniques available. The process of machine learning includes preprocessing, feature selection, building the prediction model and testing the model. In this study we have train a model to detect the Sybil accounts. Since the data is collected from the various resources so preprocessing of the dataset is done in order to remove the noisy data. Feature selection techniques are used for the selection the relevant feature. It removes the redundant and irrelevant features. In this research, we have used three feature selection methods: correlation matrix with heatmap after then feature importance and at last recursive feature elimination with cross validation. Three classifier were used to train the model. Those are random forest, support vector machine and k nearest neighbour. We have used different metrics to evaluate the results obtained from classifier. In our study we conclude that the Random Forest provides the best results out of three classifier which have been used.

Keyphrases: Classification, K-Nearest Neighbour, Preprocessing, Random Forest, feature selection

@booklet{EasyChair:2240,
  author    = {Shradha Sharma and Manu Sood},
  title     = {Exploring Feature Selection Technique in Detecting Sybil Accounts in a Social Network},
  howpublished = {EasyChair Preprint 2240},
  year      = {EasyChair, 2019}}