Causality and Its Applications

Abstract

The diagnosis of cause–effect relations boosts the both predictive and prescriptive models and their analysis. The recent trend and growth in machine learning, leading the field of predictive analytics, base the certain case for causal analysis. Causal analysis supports the study of causes and effect as they are observed and their underneath relationship directing such trends, are analysed for predictive modelling. The relation in cause–effect provides much needed information on interdependency in the various features. The interdependency helps to identify the valuable or influential features which supports for any certain trend, which is of interest. In this chapter, machine learning and necessity for causal analysis is discussed in depth. The main focus point is to provide different scenario where causal integration into machine learning is useful but lacks the process of inference. A depth view is shown for different situations where causal analysis can produce and largely effect the standard industrial process of predictive modelling. Next to this trend, a more advanced step is prescriptive analysis. Going one step further of predicting what is expected, is prescriptive analytics focus on what to avoid to achieve the required target. In this case, causal analysis can be used to find feature dependency model, leading to the prescriptive future model. As seen in machine learning modelling, much effort is put upon setting up the predictive modelling. But most industry require a better understanding of feature dependency and a model leading to business success while avoiding losses, a prescriptive modelling. In deep learning, much effort is lost on building the most accurate predictive modelling. But neural networks never seen in the lights of models which are explainable. The weights produced in most accurate models of deep learning, does not provide interpretability or their effectiveness in producing final results. With the rise of explainable AI, there is now efforts put onto the explain ability of the deep networks. This new trend/flow now opened the windows for causal explanations in deep neural network models. The task to explain deep learning models, require understanding on hidden weights and their influence on final neural outputs. Causal influence of input features through the weights inform their contribution towards neural outputs. Through explainable neural network weights, the better models can be built to overpower or underpower the effects of input features. In this note, the chapter provide how causal influence is useful and their significance for causal influence analysis. The criteria for causal influence analysis in deep neural models are discussed with examples.