Combining Error Terms into a General Error Term

Lets say I have 4 error terms:

$$e_1,e_2,e_3,e_4$$

Each of these error terms come from different simulations of data using different classification methods.

Let $gamma$ be the number of empirical data and $delta$ be the number of simulated data and $d$ is the difference of the two. The error terms are $frac{d}{gamma}$. An example is as follows (in this case the difference $d$ is always 5):

$$gamma = 500, delta = 505 implies e_1 = frac{5}{500} =1% $$
$$gamma = 50, delta = 45 implies e_2 = frac{5}{50} = 10% $$
$$gamma = 30, delta = 35 implies e_3 = frac{5}{30} =16% $$
$$gamma = 1, delta = 6 implies e_4 = frac{5}{1} = 500% $$

I want to combine these error terms into one general error term, however I want to make an emphasis on certain error terms. Because, using the average of these terms would make an error result that is too susceptible to outliers. $e_3$ is the most accurate in my opinion. Is it suitable to create a linear combination of these error terms to define a general error term?

If so, it would look like this:
$$ E = alpha_1 e_1 + alpha_2 e_2 + alpha_3 e_3 + alpha_4 e_4 , forall alpha_iinmathbb{R}$$

If this is allowable, what would be the best way to define alpha values s.t. $0<=E<=1$ OR $0%<=E<=100%$ ?

(Note: This is not the same as using a regression model).