JKRW Link Prediction – A New Ensemble Technique Based on Merging Other Known Techniques in The Social Network Analysis

Aya Taleb,Rizik M H Al-Sayyed,Hamed S Al-Bdour

doi:10.3991/ijim.v15i12.22831

Aya Taleb, Rizik M H Al-Sayyed + Show 1 more

Open Access

https://doi.org/10.3991/ijim.v15i12.22831

Copy DOI

Abstract

In this research, a new technique to improve the accuracy of the link prediction for most of the networks is proposed; it is based on the prediction ensemble approach using the voting merging technique. The new proposed ensemble called Jaccard, Katz, and Random models Wrapper (JKRW), it scales up the prediction accuracy and provides better predictions for different sizes of populations including small, medium, and large data. The proposed model has been tested and evaluated based on the area under curve (AUC) and accuracy (ACC) measures. These measures applied to the other models used in this study that has been built based on the Jaccard Coefficient, Katz, Adamic/Adar, and Preferential attachment. Results from applying the evaluation matrices verify the improvement of JKRW effectiveness and stability in comparison to the other tested models. The results from applying the Wilcoxon signed-rank method (one of the non-parametric paired tests) indicate that JKRW has significant differences compared to the other models in the different populations at <strong>0.95</strong> confident interval.

Highlights

Introduction and Background1.1 IntroductionLinks between the different networks are the most important factor to predict the generation of any of the future transactions, common interests, future friendships, communications, and many other fields
These results have led to start thinking about some links between the accuracy and population size and test size compared to the ensemble results for each population case
The other two factors that have been looked into when pursuing the evaluation of results and comparing it are the area under curve (AUC) and the time taken for building the models

Summary

Introduction

Introduction and Background1.1 IntroductionLinks between the different networks are the most important factor to predict the generation of any of the future transactions, common interests, future friendships, communications, and many other fields. The most interesting part is the question that has raised, how to properly use the magic of these network links in predicting future links in the most optimal, fast, and accurate way possible to start building facts and businesses based on these predictions. This question is one of the reasons for doing this research study. Different link prediction techniques and algorithms have been introduced to analyze the current networks and try to predict the possibility of future links and relations between the disconnected nodes in the different networks. The main characteristic of these networks is having it as a real collection of nodes and links that are dramatically increasing in size and shape over time to produce dynamic and complex networks [3]

Objectives

Methods

Results

Conclusion