Automated Pain Assessment: Is it Useful to Combine Person-Specific Data Samples?

Many classification tasks, from the field of automated pain assessment, include a special type of human-centred data sets. Often, the data is organised in person-specific data subsets. In the current work, we introduce a heuristic classification approach, which aims at explicitly making use of the particular structure of such databases. More precise, during the training phase, different person-and class-specific data samples are aggregated to form overlapping sample sets. The sample sets are fed to the classification models, which use the combined information. The aim is to improve each model's generalisation ability, by exploring the variety (sample combinations) of each participant's data distribution. We evaluate our proposed method, in combination with the publicly available pain-related BioVid Heat Pain Database. Moreover, we analyse our proposed approach in combination with three types of basic classification models, namely a nearest neighbour classifier, a support vector machine, and an ensemble of decision trees. The presented results show that our proposed approach is able to significantly improve the baseline classification performance, in combination with the decision tree based ensemble and the nearest neighbour classification model. Based on the presented outcomes, we discuss the effectiveness and potential of our proposed heuristic method, and provide promising approaches for future research directions.