produce an equation for this { 20702 20709 20695 20703} to produce this {20714}

Its not the solution to your problem, but maybe, this code snippet helps you nonetheless:

import numpy as np
import pandas as pd
df = pd.read_excel("data/data.xlsx")
df["Result"] = (df["A"] + df["B"] + df["C"] + df["D"]) // 4
df

you could think about it and try to find a pattern how the result depend on the columns (A,B,C,D)

how ever if this didn't success , there is a lot of solutions to these kind of problems in machine learning, so one could assume a linear relation between the features ( columns A,B,C,D) and the results , and try to find the parameter of an equation

Ax + By + Cz + Dw + M = result

and there's a lot more algorithm to use, if you knew about machine learning field .

A typical method is to assume that the value in Result can be obtained from the four values in A, B, C, D through some formula that involve parameters; then use an algorithm to find the parameters that best fit your data. The formula you chose is usually called a model.

This process is called regression.

In the particular case where the model looks like a weighted sum of the parameters, then this is called linear regression. This is the easiest case and the algorithms to find the best parameters are quite simple. I insist here on the fact that the term "linear" in linear regression concerns the parameters of the model, not the values from the data. For instance, if you assume that Result can be written as a polynomial in the four variables A, B, C, D, then you can use linear regression to find the coefficients of the polynomial - the model is linear in the coefficients of the polynomial, even though it is not linear in the variables A, B, C, D.

Note that the overall success of the method depends entirely on the model you chose to begin with. There are long discussions to be made about the choice between a simple model and a complicated model, and regularization techniques to allow a complicated model if necessary, but tell the optimization algorithm that if possible you would prefer a simpler model.

I won't go into more details about regularization techniques; here is a simple code using the python module sklearn.linear_model.LinearRegression to express Result as an affine combination of A, B, C, D.

import sklearn.linear_model as sklin
import pandas as pd

data = pd.read_csv('data.csv')

lm = sklin.LinearRegression()
lm.fit(data[['A', 'B', 'C', 'D']], data['Result'])

print(lm.coef_)
# array([ 0.1145072 ,  0.47290074, -0.36769957,  0.74233087])
print(lm.intercept_)
# 774.4947813684666

data['Predicted'] = lm.predict(data[['A', 'B', 'C', 'D']])
print(data)
#         A      B      C      D  Result     Predicted
# 0   20689  20724  20689  20702   20703  20704.326408
# 1   20702  20709  20695  20703   20714  20697.257624
# 2   20703  20714  20700  20714   20714  20706.063777
# 3   20714  20717  20702  20714   20728  20708.006659
# 4   20713  20732  20709  20728   20717  20722.804398
# ...

What did I do? I asked sklearn.linear_model.LinearRegression to find the best parameters a, b, c, d, e such that Result ≈ a * A + b * B + c * C + d * D + e; and the answer was [a, b, c, d] = [ 0.1145, 0.4729, -0.3677, 0.7423] and e = 774.49. I added the predictions as an extra column in the dataframe; the relation is:

Predicted = 0.1145 * A + 0.4729 * B - 0.3677 * C + 0.7423 * D + 774.49

Can we do better? You can try with a more complicated model, and see if you get better predictions.