The last algorithm designed which obtained a score of 6,4 fully used the tactic of decision trees which has an objective that matches 100% to the projects description and therefore obtained a great score.
The new proposal uses a similar approach:
1. I will train a Linear Regression over the 10 days data
2. I will use the Linear Regression to select a sub set of the data set which will be a sub set where the Linear Regression predicts a positive score
3. Finally over this sub set I will generate the query just as in the first algorithm