Using the out-of-sample forecast evaluation approach, the forecasts made by the developed and compared models when applied on the three accidents historical data sets previously employed in Sect.
The out-of-sample forecast evaluation exploits the single forecast horizon capability of GFMAPR using a recursive rolling mechanism.
Model performances were investigated on two fronts, namely accuracy of the developed set creation and partitioning method which was measured using the in-sample (trained model) prediction evaluation, and the forecast capability of the model which was investigated using the out-of-sample forecast method.
The in-fit-sample and out-of-sample forecast performance of the model was investigated in comparison with some forecasting models frequently employed in the industry by their application on three real-time industrial accidents related data.
The figures reports in-sample one period ahead probability forecasts and out-of-sample forecasts up to August 2012.
Table 3 shows the out-of-sample forecast results using linear and nonlinear time series methods.
In overcoming these shortcomings, the out-of-sample forecast evaluation method was adopted for model evaluation and validation.
The models are compared in terms of MSE, RMSE, and MAE criteria for in-sample and out-of-sample forecast capabilities.
In Table 6, models with the lowest RMSE, is denoted in bold within each model group for the above-mentioned out-of-sample forecast horizons.
To this end, we use data from the period 1991 to 2013 and perform an out-of-sample forecast for the year 2014.
We use three large, actual real-life datasets to analyze the relevance of these additional efforts for managerial interpretation and for the out-of-sample forecast accuracy at various frequencies.
