Towards the Detection of Partial Feature Interactions

Discrete feature interactions occur when the presence of overlapping, yet conflicting, features impact the functionality of an entire system such that one or more features are unsatisfied. However, adaptive systems may mitigate uncertainty by composing features from requirements that may be partially satisfied or partially unsatisfied (i.e., satisficed). In contrast to discrete interactions, these satisficed features that are composed of satisficed requirements may be partially unsatisfied due to a feature interaction. Feature interactions, in the case of satisficement, result in a proportional trade-off between conflicting objectives rather than a complete, or discrete, failure of features or the system. In this paper, we propose a new concept, partial feature interactions, to represent the class of feature interactions that allow for satisficement trade-offs between conflicting features. We also describe a method to reduce the dimensionality of the satisficement trade-offs to be reviewed when assessing the severity of the n-way interactions. Finally, we present a method for identifying Pareto optimal trade-offs among the conflicting features using the reduced dimensionality in order to detect partial feature interactions and multiple candidate adaptation points. We apply our method to an exemplar feature interaction from the automotive domain.