Incorporation of Neighborhood Concept in Enhancing SOM Based Multi-label Classification

摘要

The self-organizing map (SOM), which is a type of neural network, helps in the exploratory phase of data mining by projecting the input data into a lower-dimensional map consisting of a grid of neurons. In recent years, SOM has also been applied for classification of data points. The prominent utility of SOM based classification is evident from the use of no labeled data during training. In this paper, a self-organizing map based algorithm is proposed to solve the multi-label classification problem, named as ML-SOM. SOM follows an unsupervised training process to learn the topological structure of the training points. At testing-phase, a testing instance can be mapped to a specific neuron in the network and it's label can be determined using the training instances mapped to that specific neuron and nearby neurons. Thus in this paper, we have considered the neighborhood information of SOM to determine the label vector of testing instances. Experiments were performed on five multi-labeled datasets and performance of the proposed system is compared with various state-of-the-art methods showing competitive performance. Results are also validated using statistical significance t-test.