Progress on Statistical Learning Systems as Data Mining Tools for the Creation of Automatic Databases in Fusion Environments

Fusion devices produce tens of thousands of discharges but only a very limited part of the collected information is analysed. The analysis of physical events requires their identification and temporal location and the generation of specialized databases in relation to these time instants. The automatic determination of precise time instants in which events happen and the automatic search for potential relevant time intervals could be made thanks to classification techniques and regression techniques. Classification and regression techniques have been used for the automatic creation of specialized databases for JET and have allowed the automatic determination of disruptive / non-disruptive character of discharges. The validation of the recognition method has been carried out with 4400 JET discharges and the global success rate has been 99.02 per cent