A machine learning-based new MVA workflow to find correlations in complex data sets applied to fracture diagnostics

When attempting to optimize well design to improve fracture effectiveness in unconventional wells for given field, it is often difficult to identify the primary drivers of various fracture characteristics due to the many existing variables such as completion parameters and geologic properties along the wellbore. Additionally, many of the completion or geologic variables may have a nonlinear relationship with the performance of the treatment. This makes finding statistically significant correlations between fracture properties and different variables infeasible using traditional methods. However, in this work, we propose and evaluate a new multivariate analysis (MVA) workflow that utilizes data mining and feature selection techniques for the purpose of identifying and ranking correlations in high-dimensional, nonlinear datasets.