I prefer using soft computing methods in geotechnical problems. Because in most geotechnical problems we have a lot of data that we need to optimize them in different methods. For example, in the core of Embankment dams, we can use different algorithms to optimize its dimensions.
It depends on the problem one try to solve. Soft computing will be the choice if exact results are not required and approximation or nearby exact results are accepted. i.e. not optimum but near to optimum.
Soft computing is soft on reporting results, so if problem is complex and near to optimal solution are acceptable , in other words if can accept reporting of 'a' solution but not very stringent on getting a particular 'the' solution , then soft computing gives a easy and natural way out.
But if our problem domain strictly need to find out the perfect optimal solution, and near to optimal are not acceptable , then hard computing.
In situations where both can be applied , then the choice is entirely dependent on the execution complexity , dataset size. For hard computing proofs are justified but soft computing is entirely stocastic.
Complex problems, wherein many real-world problems fall into this category, are generally characterized by one or more of the following bewildering features: 1) Incomplete and limited domain knowledge. 2) Ambiguous and fuzzy information gathered from the problem. 3) Non-linear interaction among the parameters of the problem. 4) Contradictory objectives 5) Instability and dynamic environment within which the problem lives. To solve a black-box problem with such non-deterministic features more practically, reasonably, adaptively, and efficiently, we need to identify some flexible problem-solving techniques that can work softly to search for the required needle solution in a haystack deceptive space rather than stopping to ask us to hardly determine the steps to find the solution (by which we need to convert the problem to a white-box one and let the algorithm to search for that required needle!!). The majority of soft computing (SC) has, in the literature, three main streams: Nature-inspired computing (NIC), Fuzzy computing (FC), and Neural computing (NC).