Oil Flow Predictor (Barrels Per Day) Using Machine Learning Algorithms – Random Forest and Gradient Boosting 
Machine. Executives in the Oil & Gas industry are always seeking ways to optimize the production of oil and gas 
wells and make informed decisions about drilling and exploration. Machine learning can be used to effectively 
resolve this optimization problem, but would usually involve extensive or complex coding that the typical 
executive or non-technical staff would not be able to undertake. For example, the code behind this simple Gradio
bot does thinngs the typical oil executive knows nothing about in that it loads the data from a CSV file and 
performs data scaling, PCA dimensionality reduction, DBSCAN clustering, and random forest regression and
gradient boosting regression for predicting output variables related to oil and gas production rates – not 
something the typical executive or non-technical Oil & Gas staff can manage. The analysis was used to gain 
insights into the factors that affect oil and gas production rates and to predict production rates based on 
the input variables. Executives are then able to use this knowledge to improve the efficiency of their 
operations. 

This Gradio chatbot reads in a CSV file containing well rate data, performs some preprocessing on the data, and
then trains and evaluates two regression models, a Random Forest (RF) and a Gradient Boosting Machine (GBM), on 
the preprocessed data. The input data is preprocessed using Principal Component Analysis (PCA) to reduce the 
dimensionality of the input features, and then the DBSCAN clustering algorithm is applied to the PCA-transformed
data to filter out outlier data points. The remaining data points are used for training and testing the
regression models. The RF and GBM models are both trained using the preprocessed input features and 
corresponding well rates, and their performance is evaluated using the R-squared metric. Finally, the predicted
well rates from the RF and GBM models are plotted against the actual well rates for visualization.

For example, the following inputs are required for predicting the oil flow rate (Qoil) based on input features 
(BHP, WHP, WHT, Tsep, Psep, and Choke_in):

•	BHP: Bottom Hole Pressure
•	WHP: Well Head Pressure
•	WHT: Well Head Temperature
•	Tsep: Separator Temperature
•	Psep: Separator Pressure
•	Choke_in: Choke Size

These features (BHP, WHP, WHT, Tsep, Psep, and Choke_in) are used as inputs to the machine learning models that
were trained in the notebook to predict the oil flow rate (Qoil). We suspect and propose that the same input
features (BHP, WHP, WHT, Tsep, Psep, and Choke_in) are provided to the trained model which would then be 
enabled to predict Qoil or the oil flow rate (number of barrels per day) via the Gradio input interface.
The oil executive is able to learn this without having to know machine learning or the complexities of 
training the model himself.