Prediction of the CATS benchmark using a business forecasting approach to multilayer perceptron modelling

Various heuristic approaches have been proposed to limit design complexity and computing time in artificial neural network modelling and parameterisation for time series prediction, with no single approach demonstrating robust superiority on arbitrary datasets. In business forecasting competitions, simple methods robustly outperform complex methods and expert teams. To reflect this, we follow a simple neural network modelling approach, utilising linear autoregressive lags and an extensive enumeration of important modelling parameters, effectively modelling a miniature forecasting competition. Experimental predictions are computed for the CATS benchmark using a standard multilayer perceptron to predict 100 missing values in five datasets.