Procedure to Select the Best Dataset for a Task

This paper models the decision process when selecting among different datasets the one most suitable for a task. It shows how metadata describing the quality of the dataset and descriptions of the task are used to make this decision. A simple comparison of task requirements and available data quality is supplemented with general, common-sense knowledge about effects of errors, lack of precision in the data and the dilution of quality over time. It consists of two steps: first, compute the data quality considering the time elapsed since the data collection; and second, assess the utility of the available data for the decision. A practical example of an assessment of the suitability of two datasets for two different tasks is computed and leads to the intuitively expected result.