Neural Learning from Unbalanced Data Using Noise Modeling

This paper describes the result of our study on neural learning to solve the classification problems in which data is unbalanced and noisy. We use multidimensional Gaussian distribution to analyze the separation of different class samples in a training data set, and then generate artificial noise samples in the training set using a noise modeling algorithm. The noise analysis allows us to identify special densities in the feature space that are prone to prediction error. We argue that by properly generate extra training data samples around the noise densities, we can train a neural network to have stronger capability of generalization and control the classification error of the trained neural network. In particular, we focus on the problems that require a neural network to make favorable classification to a particular class. The noise modeling algorithm has been implemented to solve a classification problem of good(pass) and bad(fail) vehicles in test sites of automobile assembly plants and a multi-layered Back Propagation neural network has been used in our experiments. The experimental results showed that the noise modeling algorithm was very effective in generate extra data samples that can be used to train a neural network to make favorable decisions to a minority class and to have increased generalization