Link Prediction Using Evolutionary Neural Network Models
Author(s)
Yaghi, Rawan I
Faris, Hossam
Aljarah, Ibrahim
Al-Zoubi, Ala’ M
Heidari, Ali Asghar
Mirjalili, Seyedali
Griffith University Author(s)
Year published
2020
Metadata
Show full item recordAbstract
Link prediction aims to represent the dynamic networks’ relationships of the real world in a model for predicting future links or relationships. This model can help in understanding the evolution of interactions and relationships between network members. Many applications use link prediction such as recommendation systems. Most of the existing link prediction algorithms are based on similarity measures, such as common neighbors and the Adamic/Adar index. The main disadvantage of these algorithms is the low accuracy of results since they depend on the application domain. Moreover, the datasets of link prediction have two ...
View more >Link prediction aims to represent the dynamic networks’ relationships of the real world in a model for predicting future links or relationships. This model can help in understanding the evolution of interactions and relationships between network members. Many applications use link prediction such as recommendation systems. Most of the existing link prediction algorithms are based on similarity measures, such as common neighbors and the Adamic/Adar index. The main disadvantage of these algorithms is the low accuracy of results since they depend on the application domain. Moreover, the datasets of link prediction have two significant problems: the imbalanced class distribution and the large size of the data. In this chapter, evolutionary neural network-based models are developed to solve this problem. Three optimizers are used for training feedforward neural network models including genetic algorithm, particle swarm optimization, and moth search. For this purpose, the link prediction problem is formulated as a classification problem to improve the accuracy of the results by constructing features of the traditional link prediction methods and centrality measures in any given link prediction dataset. Also, this work tries to address two problems of the data in two ways: externally using sampling techniques (random and undersampling) and internally using the geometric mean as a fitness function in the proposed algorithms. The results reveal that the proposed model is superior in terms of the sensitivity and geometric mean measures compared to the traditional classifiers and traditional link prediction algorithms.
View less >
View more >Link prediction aims to represent the dynamic networks’ relationships of the real world in a model for predicting future links or relationships. This model can help in understanding the evolution of interactions and relationships between network members. Many applications use link prediction such as recommendation systems. Most of the existing link prediction algorithms are based on similarity measures, such as common neighbors and the Adamic/Adar index. The main disadvantage of these algorithms is the low accuracy of results since they depend on the application domain. Moreover, the datasets of link prediction have two significant problems: the imbalanced class distribution and the large size of the data. In this chapter, evolutionary neural network-based models are developed to solve this problem. Three optimizers are used for training feedforward neural network models including genetic algorithm, particle swarm optimization, and moth search. For this purpose, the link prediction problem is formulated as a classification problem to improve the accuracy of the results by constructing features of the traditional link prediction methods and centrality measures in any given link prediction dataset. Also, this work tries to address two problems of the data in two ways: externally using sampling techniques (random and undersampling) and internally using the geometric mean as a fitness function in the proposed algorithms. The results reveal that the proposed model is superior in terms of the sensitivity and geometric mean measures compared to the traditional classifiers and traditional link prediction algorithms.
View less >
Book Title
Evolutionary Machine Learning Techniques
Subject
Neural networks
Machine learning
Artificial intelligence