A distributed approach for supervised som and application to facies classification

Abstract

This study proposes a distributed classification framework, which adapts supervised SelfOrganizing Maps (SOM) as base learners. The supervised SOM is the integration of the SOM algorithm with the Learning Vector Quantization (LVQ) algorithm, so called SOM-LVQ model. Multiple SOM-LVQ models are created using different feature subsets, each of which represents one different local information source. This
approach aims at utilizing the information hidden in smaller feature subsets, that cannot be obtained if the data is processed as a whole original feature combination. The outputs from all different local supervised SOMs are fused together using some specific fusing rule to provide the
final decision on the class label of the input data. This proposed distributed classification approach is applied on well-log data to determine the facies classes of the log samples. Experiments are conducted based on the well-log data-set collected from Cuu Long basin, which is an early
Tertiary rift basin located off the southeast coast of Vietnam. The experimental results show that the newly proposed distributed supervised SOM-based classification approach outperforms not only the single supervised SOM model but also some other commonly used machine
learning models in terms of accuracy rate. It is also shown that the distributed approach is more useful when the number of input features is high, and is a flexible solution for many real-life classification problems.