Resting state functional connectivity is one of the key factors in studying the cognitive characteristics in patients with Autism Spectrum Disorder (ASD). Graph theoretic measures involving centrality, path length and nodal/global efficiency are calculated to study these network characteristics. In this study, ABIDE open dataset is used, constituting ASD and TD participants from CMU and Caltech sites. Centrality measures involving degree centrality, betweenness centrality, clustering coefficient and eigenvector centrality are measured at different thresholds of network connectivity. Linear SVM is used to distinguish between ASD and TD using the above measures. Experimental results show that among all measures, clustering coefficient showed a higher accuracy of 79.17% with specificity and sensitivity of 75.0% and 83.3% respectively for participants from CMU site at 0.75 threshold. Although the overall classifier performance is lower in all measures across threshold levels, sensitivity with respect to clustering coefficient and eigenvector centrality has shown an increase with the increase in network sparsity. As the threshold values are increased, i.e., with a sparser network representation the specificity improves in a site-specific classifier, while the same is not observed on another site probably due to inter-site variability.