Hello everyone and thank you for reading my question.
I have a data set that have around 2000 data point. It have 5 inputs (4 wells rate and the 5th is the time) and 2 ouputs ( oil cumulative and water cumulative). See the attached image.
I want to build a Proxy model to simualte the cumulative oil & water.
I have made 5 models ( ANN, Extrem Gradient Boost, Gradient Boost, Randam forest, SVM) and i have used GridSearch to tune the hyper parameters and the results for training the models are good. Of course I have spilited the training data set to training, test and validation sets.
So I have another data that I haven't include in either of the train,test and validation sets and when I use the models to predict the output for this data set the models results are bad ( failed to predict).
I think the problem lies in the data itself because the only input parameter that changes are the (days) parameter while the other remains constant.
But the problem is I can't remove the well rate or join them into a single variable because after the Proxy model has been made I want to optimize the well rates to maximize oil and minimize water cumulative respectively.
Is there a solution to suchlike issue?
hi Mohammed , you right
if the input is constant that mean have no effect on output for example
linear regression model :
y = a1* x1 + a2 * x2 + a3
if x1 constant so parameter a1 = 0 Because the model is trained on data.
There is no solution except to use more data with many rates in same well
To everyone who faced this problem, this type of data is called time series data which have a specific algorithm that used to build the proxy models (i.e RNN, LSTM)