Usually, I think the goal of machine learning is to minimize the discrepancy between the actual value and the predicted value of the objective variable.
The question I would like to ask this time is whether it is possible to solve the problem by machine learning so that the difference between the numerical value calculated based on the estimation result of the objective variable and a certain numerical value is minimized.
The image is as follows.
If there is an existing machine learning model that solves such a problem, I would appreciate it if you could teach me about that model.

  • Answer # 1

    I think you should use a neural network.
    In the case of the question, define the predicted value as a hidden layer so that it is calculated from the explanatory variables. Next, calculate the final desired value from the predicted value using the specified formula.
    Now that the body of the model is complete, I think we should define a cost function that evaluates the difference from the objective variable and find the parameter that minimizes this cost function (the coefficient that calculates the hidden layer from the explanatory variables).

  • Answer # 2

    I think the essence of the problem is the same, with only 18 explanatory variables. Since there is a correlation between the explanatory variables, it is not easy to predict the increase.

    In other words, we used to predict the monthly average temperature every month, but it is equivalent to predicting the average temperature every three months using the same explanatory variables. think.