For decades, geoscientists have employed seismic attributes to improve their understanding of the subsurface. More recently, image-based texture attributes started to be considered in the geoscience community to describe the texture patterns found in seismic data. In this context, many works have investigated seismic and texture attributes and their combination for different tasks in seismic interpretation. This work aims to analyze the performance of seismic and texture attributes for a machine learning task from a feature selection perspective. We select the most discriminative attributes for a classification and a clustering task related to the stratigraphic segmentation of a public seismic dataset. The results indicate that texture attributes may be more suitable for these tasks than seismic attributes. Ten out of ~60 attributes selected with the ANOVA feature selection algorithm achieved 0.85 of F1 Score and 0.67 of V-Measure for the classification and clustering tasks, respectively.