Fraud detection and prevention systems are based on various technological paradigms but the most prevailing one is rule-based reasoning. However, most of the existing rule-based fraud detection systems consist of fixed and inflexible decision-making rules which limit significantly the effectiveness of such systems. In this paper we present a fraud detection approach which combines the technologies of knowledge-based systems and adaptive systems in order to overcome the limitations of traditional rule-based reasoning. Our approach is supported by an integrated generic methodology for addressing fraud in various e-government domains and organizations through a number of well defined steps that ensure the efficient application of the approach. It is supported also by a generic ontological framework based on which different domain specific fraud knowledge models can be built and through which the generic character and adaptability of our approach is ensured.