The detection of ischemic episodes is a difficult pattern classification problem. The motivation for developing the Supervising Network - Self Organizing Map (sNet-SOM) model is to design computationally effective solutions for the particular problem of ischemia detection and other similar applications. The sNet-SOM uses unsupervised learning for the regions where the classification is not ambiguous and supervised for the 'difficult' ones- in a two-stage learning process. The unsupervised learning approach extends and adapts the Self Organizing Map (SOM) algorithm of Kohonen. The basic SOM is modified with a dynamic expansion process controlled with an entropy based criterion that allows the adaptive formation of the proper SOM structure. This extension proceeds until the total number of training patterns that are mapped to neurons with high entropy (therefore with ambiguous classification) reduces to a size manageable numerically with a proper supervised model. The second learning phase (supervised training) has the objective of constructing better decision boundaries of the ambiguous regions. In this phase, a special supervised network is trained for the task of reduced computationally complexity to perform the classification only of the ambiguous regions. After we tried with different classes of supervised networks, we obtained the best results with the Support Vector Machines (SVM) as local experts. Keywords - Self-Organizing Maps, Ischemia, Entropy, Principal Component Analysis, Divide and Conquer algorithms, Radial Basis Functions, Vapnik-Chervonenkis Dimension, Support Vector Machines, Computational Complexity.
展开▼
