| Input: ParticipantSensitiveness data from G00, G01, G10, G11 |
| Output: Build indicators’ record T with decision tree model |
| (1) for each group Gxy do // classify each user group |
| (2) train dataset Answers by 10-folds-validation |
| (3) to predict indicators (Ic, Id) //train the data by turns |
| (4) Create a point Root //start to build the tree |
| (5) if all the participants in Gxy belong to one class C |
| (6) return Root as a single leaf, label as C //all users are same |
| (7) Else find best splits of subclasses C |
| (8) with highest prediction accuracy P //find the best splits |
| (9) repeat find further splits for each subclass cx |
| (10) if find higher prediction accuracy |
| (11) return to step (9) //for more specified splits |
| (12) else return the value record of (Ic, Id) |
| (13) T = T + (Ic, Id) // add leaf to the tree model |
| (14) if Answers ≠ //add all leaves to the tree |
| (15) Answers = Answers – RequestedItemX |
| (16) where RequestedItemX is the last column of Answers |
| (17) return to step (2) // the tree model is built |
| (18) until Answers = |
