删除所有观察值具有相同值的列会影响我的模型吗？

Question

One of the columns in my dataset has the same value for all observations/rows. Should I remove that column while building a machine learning model?

Will removing this column affect my model/performance metric?

If I replace all the values with a different constant value, will it change the model/performance metric?

Answer 1

If one of your column in the dataset is having the same values, you can drop this column as it will not do any help to your model to differentiate between two different labels while on the other hand, it can even negatively affect your model by creating a bias in the data.

For Example: Consider you have two different fruits, like one is Green Apple and one is Guava. Then, both of these fruits will have the same color ie "Green", so that basically means that you just can not differentiate both these fruits on the basis of their color, but if they have been two different colored fruits, you could have used this feature to differentiate between them.

Hope it helps clarifying what you should do with such a column with same set of observations.

Thanks.

Answer 2

A Machine Learning Model is nothing but a mathematical equation ie

y = f(x)

in which

y = Target/Dependent Variable

f(x) = Independent Variables(In our case a DataFrame containing the Train/Test Data)

So technically, ML models quantifies and estimates about for what value of X, what will the probable output y.

Assuming a single whole column is constant. So, a relationship between y and f(x=constant) is meaningless because for whatever value of y, that x will remain same. No mathematical relationship is possible except for the only option that y is also an constant. Which we can safely assume isn't the case, else why else you will build a model to get a constant value.

Hence, we can safely drop any constant column, which doesn't add any variation in data to the DataFrame to save computational time, as that column won't affect y in any sense.