msdt2-multi-valued-ExpII.accdb

The msdt2-multi-valued dataset for Exp. II has three large multi-valued and sequential-labeled datasets with α = 3..5. We apply the MSDT algorithm on handling a large dataset of tourists. The large dataset has the same features and values as the dataset used by both the multi-valued and multi-labeled algorithms of MMC and MMDT, except it has two additional attributes, one for item-sequences and the other for sequential-label. And, three features of the dataset are multi-valued. To learn a classifier from incrementally-added sequences of training instances within some successive sliding windows per classification lifecycle, we set the user-specified parameters that the sliding window sizes, α = 3..5. Therefore, the value of the “sequences” attribute is a set of item-sequences with lengths from 3 to 5, which correspond three most recent sliding windows respectively with incrementally-added sizes. We thus have three large datasets of 3-sequence, 4-sequence and 5-sequence. The item-sequences of each record are generated by the combination of the five functions, which can be accessed at our data repository over the Harvard Dataverse repository. The number of various sequential-labels is 10 while α = 3; 13 while α = 4; and 10 while α = 5. All the experiments use 5,000 records as the test dataset. The size of 8 training-set are 90, 1,000, 4,000, 6,000, 8,000, 10,000, 12,000 and 14,000 records respectively.