Example implementations described herein create a rich feature set based on observed/recorded attributes as well as attributes derived from those, and models each well as a data vector in this multi-dimensional attribute space. Example implementations then compute composite similarity between wells which provides better insights into their behavior. This composite similarity can be calculated along all dimensions or subsets of dimensions, and serve as input to any clustering algorithm for further analysis. Finally, the similarity can be incrementally computed by incorporating more attributes as required. Such implementations can be used to provide insights into behavior of oil wells, especially horizontal wells by integrating features from multiple upstream processes.