Generalizing, Decoding, and Optimizing Support Vector Machine Classification

The classification of complex data usually requires the composition of processing steps. Here, a major challenge is the selection of optimal algorithms for preprocessing and classification. Nowadays, parts of the optimization process are automized but expert knowledge and manual work are still required. We present three steps to face this process and ease the optimization. Namely, we take a theoretical view on classical classifiers, provide an approach to interpret the classifier together with the preprocessing, and integrate both into one framework which enables a semiautomatic optimization of the processing chain and which interfaces numerous algorithms.

First, we summarize the connections between support vector machine variants (SVM) variants and introduce a generalized model which shows that these variants are not to be taken separately but that they are highly connected. The model improves the understanding of relationships between the variants. It can be used to improve teaching and to facilitate the choice and implementation of the classifiers. Often, knowledge about and implementations of one classifier can be transferred to the variants. Furthermore, the connections also reveal possible problems when applying some variants.

Having complex, high dimensional data and consequently a more complex processing chain as a concatenation of different algorithms, up to now it was nearly impossible to find out what happened in the classification process and which components of the original data were used. So in our second step, we introduce an approach called backtransformation. It enables a visualization of the complete processing chain in the input data space and thereby allows for a joint interpretation of preprocessing and classification to decode the decision process.

The third step is meant for the practitioner and hence a bit more technical. We propose the signal processing and classification environment pySPACE which enables the systematic evaluation and comparison of algorithms. It makes the aforementioned approaches usable for the public. Furthermore it enables the combination of approaches and provides an interface to optimize the processing chain.