Autism Spectrum Disorders (ASD) is an array of neurodevelopmental disorders that are characterized by social and communicative deficits and repetitive behavior. ASD has been reported to affect approximately 2% of the population. As the name implies, ASD is characterized by great heterogeneity of symptoms. This has led to behavioral and genetic studies that have attempted to characterize the etiological subtypes of ASD. Characterizing the subtypes can potentially lead to more personalized and effective interventions for individuals with ASD. Although ASD has been studied extensively by means of resting state functional MRI (rs-FMRI) data, few studies have focused on imaging data with the goal of characterizing subtypes of ASD. The present study combined unsupervised feature selection methods with data-driven cluster analyses to explore the possible subtypes of ASD in rs-fMRI data. The effects of regressing nuisance variables and preprocessing with global signal regression are compared and observed. Optimal clustering solutions varied greatly between data processing procedures. The two-cluster solution from the nuisance variable regressed data disappeared when the data were preprocessed with global signal regression. The inconsistent results can be attributed to a slew of potential problems. Although the results of the study were not clinically meaningful, many suggestions can be made to improve the efficacy of future studies. Similar future studies should consider better quality control of data, subspace clustering algorithms, and semi-supervised feature selection.