Graph connectedness and Homophily: An inquiry into feature selection

Homophily in graphs can be well understood if the underlying causes for homophilic associations are identified. As associations are generally based on feature selection, it necessitates understanding the reasons for decision making for feature selection. This work tries to establish relationships between graphs’ properties and the degrees of homophily exhibited by them on the basis of feature selection approaches which entails decision-making. The analysis is performed in both social and non-social context using three approaches for feature selection based on importance, variance within features and dimensionality reduction. This work by co-relating the properties of datasets and homophily across different feature selection approaches and different dataset sizes, proves that the importance-based feature selection is the most likely approach for feature selection for homophilic associations in graphs.