Application of Evolutionary Algorithms in Social Networks: A Comparative Machine Learning Perspective

Social networks exhibit interactions that lead to event changes in their communities. It is imperative to track community events to understand an extensive social network. Recently, several models reported that the randomness and sparsity of social networks bring significant challenges in predicting community events. Hence, the proposed work extracts both community and temporal features to predict the events effectively that a community might experience. Machine learning (ML) models are widely used in predicting such events in a social network. Many machine learning models, such as naive Bayes, random forest, logistic regression, SVM, neural networks, etc., are used to predict community events. Further, the model’s performance is enhanced using hyper-parameter tuning by selecting the appropriate parameters. Evolutionary algorithms are effective in tuning these hyper-parameters. This paper investigates the effectiveness of Cuckoo search optimization (CSO), particle swarm optimization (PSO), ant colony optimization (ACO), jellyfish search optimization (JFO), and mayfly optimization (MFO) evolutionary algorithms in tuning the hyper-parameters of four ML models to achieve higher accuracy in the results. The comparative analysis of these 20 combinations (five evolutionary algorithms and four ML models) shows that CSO improves average accuracy by 4.12% in all the machine learning models compared to PSO, ACO, JFO, and MFO. Furthermore, results confirm that CSO precisely suits the neural network model in tuning its hyper-parameters. The accuracy of the neural network model improved by 4.5% after tuning its hyper-parameters using CSO.