Modelling Conflict: Knowledge Extraction using Bayesian Neural Network and Neuro-fuzzy Models

Much has been written about the lack of transparency of computational intelligence models. This paper investigates the level of transparency of the Takagi-Sugeno neuro-fuzzy model and the Neural Network model by applying them to conflict management, an application which is concerned with causal interpretations of results. The neural network model is trained using the Bayesian framework. It is found that the neural network is able to forecast conflict with an accuracy of 77.30%. Knowledge from the neural network model is then extracted using the Automatic Relevance Determination method and by performing a sensitivity analyis. The Takagi-Sugeno Neuro-fuzzy model is optimised to forecast conflict giving an accuracy 80.36%. Knowledge from the Takagi-Sugeno neuro-fuzzy model is extracted by interpreting the model’s fuzzy rules and their outcomes. It is found that both models offer some transparency which helps in understanding conflict management.