WELL PREDICTIONS(Chevron Oil Peak Rate Prediction)

Inspiration

Addressing the complexity of predicting peak oil production rates, our initial emphasis was on data cleaning. This process included eliminating columns with over 80% missing data and replaced other column null values by using imputation methods. Our initial strategy involved constructing a CatBoostRegressor to gain insights into the dataset. By running cleaning dataset on different machine learning models we compared using RMSE values

What it does

Given a dataset our goal was to find a model that predicts the Oil Peak Rate

How we built it

The initial phase involved data cleaning, where we analyzed variability and removed outliers based on the analysis. Subsequently, we conducted correlation analysis to determine the variables for our model. For the columns with null values we used different imputation strategies. We proceeded to train and test multiple models, including KNN,Decision Tree, Random Forest, CatBoostRegressor, SVM RGB Kernal, and XGboost.To deal with the target column null values initially we divided into train which has no null values and test with null values and by running different machine leaning on those data,able to recognize best mode and finding the missing values that is test that and impute those values in the original data. After thorough hyperparameter tuning, we ultimately selected XGboostRegressor as our chosen model.

Challenges we ran into

The major challenge is handling null values as there are many columns with null values including the target column, dividing which variables were most important, and then tuning parameters to get the best fit model.

Accomplishments that we're proud of

Successfully handled the missing data including target column data and we were able to reduce the RMSE of our training and validation data throughout the project.

What we learned

We are not aware of oil and gas domain knowledge. By doing this chevron track we learnt different terminologies in this domain and handling of missing values in target variable which we never worked before