Probability density function estimation based on representative data samples

The relationship between the results of probability density function (PDF) estimation based on Parzen windows method and the number of observed samples is demonstrated in this paper. Based on the experimental analysis, we get that the increase of observed samples may not bring about the obvious improvement of estimated result. Then, the strategy by using the representative data samples to estimate PDF is proposed. The representative data samples are selected from the original dataset by considering Entropy-Maximization and Distance-Minimization (EMDM). Finally, the experimental results on the artificial datasets shows that the estimations of PDF by using the representative data samples can obtain the similar levels of error performance compared with the estimations on the whole dataset.