Configuration parameter name-value pairs are obtained for each of a plurality of component instances in a deployed software solution. A candidate set of configuration dependencies between different ones of the plurality of component instances in the deployed software solution is identified, based on the configuration parameter name-value pairs. The candidate set of configuration dependencies includes true dependencies and false dependencies. The candidate set of configuration dependencies is rank-ordered to obtain a rank-ordered list, such that a probability of the true dependencies getting a higher rank is greater than a probability of the false dependencies getting a higher rank.