O wzbogacaniu metod klasyfikacji w zdolność do wyrażania wątpliwości w przypadkach trudnych do rozstrzygnięcia
Providing classification methods with the ability to hesitate in cases difficult to solve

Summary/Abstract: When making decision sometimes we hesitate, because we feel it can go both directions. The situation is similar in machine-learning tasks. We can use very sophisticated classification methods to support our decision-making process. The machine is learned, the model is built, but then it seems reasonable to expect the machine to give us at least a warning when the prediction is unstable (which means that it is sensitive to small changes in explanatory variables’ values). The main goal of the article is to present a procedure for providing the machine with the ability to show hesitation, when it is justified. The proposed procedure is based on sensitivity analysis. We illustrate the procedure on a real-world data set using the Support Vector Machines, but the procedure is universal and it can be also used with other classification methods. The added value of the paper is also the proposed type of plot for visualizing the outcome of the sensitivity analysis.

• Details
• Contents