Thank you for your comments Edson.
You are absolutely right in what you say. However, the absence of date as reference in calculating head to head statistics has several reasons.
Firstly, as you can see in the description of 'H2H' statistics, the low statistics are filtered. This shows that the variable 'H2H' variable initially was not intended to change over time, rather I wanted to have a general indicator of head to head clashes over some relative time period.
Secondly, adding date as a reference has a big drawback. Actually it does not do what it is supposed to do. It does not show the 'past' performance in absolute manner (as our data does not contain all the historical data), but shows the 'past' performance of two particular teams only relative to the time period in the dataset that we give to the classifier.
And I think the latter is the reason why adding the date as reference decreases the accuracy of the model. So, the following sentence in the text:
''It [H2H] basically illustrates the advantage of Team1 over Team2 during their previous matches.''
means the previous matches before the match that is being predicted. (I agree it is not well formulated).
