Well, Well, (Oil) Well

Inspiration

Oil companies need to make strategic decisions based on oil rates

What it does

Predicts Oil Peak Rates

How we built it

Data Exploration & Insights, Smart imputation of missing values, Engineering new features And trying out the many models, the best of which are Logistic Regression, XGBoost, Catboost, Model Stacking

Challenges we ran into

A lot of missing data + lack of expertise in petroleum engineering

Accomplishments that we're proud of

Very low MSE, a generalizable mode, and finding the most important features! Engineer more features: Given the feature importance findings, better data collection and further exploration of additional imputation methods is valuable. Model Stacking We’d like to start observing the angle of the well decline The importance of true vertical depth, fluid/proppant intensity, and surface-toe distance makes this clear We’d like to consider more geographical (potentially deanonymized) data to be able to cluster wells by location A deeper understanding of how Oil Wells work could provide insight for engineering new features

What we learned

Many ways of data imputations, learning the relationship in the sparse dataset, and understanding why some models behave better than others.

What's next for Well, Well, (Oil) Well

Engineer more features: Given the feature importance findings, better data collection and further exploration of additional imputation methods is valuable. We’d like to start observing the angle of the well decline The importance of true vertical depth, fluid/proppant intensity, and surface-toe distance makes this clear We’d like to consider more geographical (potentially deanonymized) data to be able to cluster wells by location A deeper understanding of how Oil Wells work could provide with insight for engineering new features