Conditional and unconditional safety performance forecasts for aviation predictive risk management

This paper deals with safety performance predictions in the aviation, which address the long-term global efforts to achieve predictive risk management by the year 2028. Predictive risk management regards timely and accurate detection of risk, well before some incident or accident takes place so that effective control actions can be provided. To assure achieving such diagnosis, it is necessary that mathematically well-founded predictions will become part of existing safety management systems with the capability to predict key performance indicators. From current safety metrics and with respect to the data available in the aviation, overall safety performance was selected as suitable candidate for predictions. To obtain the performance signal, Aerospace Performance Factor methodology was utilized. Due to confidentiality restrictions with regard to aviation safety data, this study relies on public data sets from the domain of European Air Traffic Management. Dedicated resampling method was used to fill in the gaps of real data sets by transforming expert knowledge into mathematical functions. This enabled the possibility to build and test mathematical models for predicting safety performance. Because the identified data sources included some data, which are not necessary for computing safety performance but relevant in its context, conditional forecasts were made possible. With respect to this, the goal of this paper was to research and evaluate possibilities for both conditional and unconditional forecasts in the context of future risk management. Time-series analysis of the computed safety performance was conducted using ordinary least squares and maximum likelihood estimation. Each of the methodology led to different mathematical model and different predictions. Specific aspects of each methodology were identified. Among others, the conclusions confirm possibility of predicting safety performance for establishing predictive risk management, highlighting great potential of conditional forecast and favouring systemic models of safety.