Reverse Engineering Complex Feature Correlations for Product Line Configuration Improvement

As a Software Product Line (SPL) evolves with increasing number of variant features and feature values, the feature correlations become extremely intricate. However, these correlations are often incompletely documented (e.g., In feature models) so that most features can only be configured manually. In order to make product configuration processes more efficient, we present an approach to extracting complex feature correlations from existing product configurations using association mining techniques. Then these correlations are pruned and prioritized in order to minimize the effort of correlation validation. Our approach is conducted on an industrial SPL with 100 product configurations across 480 features. While 80 out of the 100 configurations are used as training data to automatically extract 4834 complex feature correlations, the rest 20 configurations are used as test data to evaluate the improvement potential of configuration efficiency. In the end, avg. 25% features in each of the 20 products can be configured automatically.