next up previous contents
Next: Accuracy measurement Up: Practical details Previous: Practical details   Contents

Metafeatures

Since TClass is novel, in this first series of experiments, it was decided the simple set of metafeatures discussed in the previous chapter would be used (increasing, decreasing, plateau, local maximum and local minimum). This would allow us to focus on the algorithm itself and make observations about the generalisability of these simple metafeatures. Further, these simple metafeatures provide a lower bound on performance; more complex metafeatures could only potentially improve performance (otherwise, temporal classifiers could revert to simple metafeatures).



Mohammed Waleed Kadous 2002-12-10