An exponential mixture models for noise reduction in clustering

Noise is an important factor that influences the performance of clustering or classifying. Many researches on noise reduction were reported in this eld. In this paper we assume that the noise data and the real data integrate with each other, and we design an exponential mixture model to reduce noise data for improvement of performance of clustering. This model is based on the mathematic theory which is that two exponential family distributions multiply with each other to get another exponential family distribution. If the noise data is Gaussian distribution and real data is Gaussian distribution, the observed data also is Gaussian distribution. In the real case the observed data can be measured, but the real data can not be known. This exponential mixture model´s goal is to reduce the noise data from the observed data. Finally the real data can be obtained. We use this model to preprocess the observed data for clustering, we found that the performance is improved much, which shows that the model works well. Noise is an important factor that influences the performance of clustering or classifying. Many researches on noise reduction were reported in this eld. In this paper we assume that the noise data and the real data integrate with each other, and we design an exponential mixture model to reduce noise data for improvement of performance of clustering. This model is based on the mathematic theory which is that two exponential family distributions multiply with each other to get another exponential family distribution. If the noise data is Gaussian distribution and real data is Gaussian distribution, the observed data also is Gaussian distribution. In the real case the observed data can be measured, but the real data can not be known. This exponential mixture model´s goal is to reduce the noise data from the observed data. Finally the real data can be obtained. We use this model to preprocess the observed data for clustering, we found that the performance is improv- - ed much, which shows that the model works well.