Since SVM can capture geometric characteristics of feature space and capable of extracting the optimal solution with the small training set size,the classifier of SVM approach outperforms BPN to the problem of corporate bankruptcy prediction.

A new recognition method, Support Vector Machine (SVM) was presented, which has excellent learning and generalization ability in solving learning problem with small training set of sample.

With a small training set and a simple network architecture,a high prediction accuracy has been achieved,i. e. ,self-consistence accuracy 97.62%,jack-knife test accuracy 97.62% and extrapolating accuracy 90.74% on average.

However, data accumulated in manufacturing plants have unique characteristics, such as unbalanced distribution of the target attribute, and a small training set relative to the number of input features.

We demonstrate its suitability for high-dimensional problems with small training set sizes.

Using the small training set, we generated several population of 5 up to 10 MLPs.

The small training set size is selected as an exercise in exploration intended to push the limits of generalization.

The one outlier is family 2.44.1.2, which has a relatively small training set.

The statistical-fuzzy method and the fuzzy-statistical method for pattern recognition are developed in this paper on the bas's of the discussion on the generalities, differences and the respective suitable scopes of statistical approaches and fuzzy approaches to puttern recognition. The statistical-fuzzy method is to adopt in a fuzzy classifier the membership functions which make full use of the statistical information of the pattern components, so that the performance of the classifier is better than that of...

The statistical-fuzzy method and the fuzzy-statistical method for pattern recognition are developed in this paper on the bas's of the discussion on the generalities, differences and the respective suitable scopes of statistical approaches and fuzzy approaches to puttern recognition. The statistical-fuzzy method is to adopt in a fuzzy classifier the membership functions which make full use of the statistical information of the pattern components, so that the performance of the classifier is better than that of common fuzzy classifiers. The fuzzy-statistical method is to replace the pattern components by their fuzzy membership functions as inputs in a. classifier which is based on the statistical method. From the results of the classification experiments made with the data sets given in this paper, it can be seen that the classification performance of this method can approach the optimal level of the Bayesian classifier with quite a small training set.

This work presents a backpropagation neural network trained to reproduce the reaction yield of aryl fluorides by the halex technique. The work shows that a ten\|dimensional input space is able to reproduce reasonably the observed reaction yields by employing statistics in artificial neural system. By means of a number of multilayer feedforward (MLF) networks rather than one, the disadvantages caused by network randomness are limited greatly, and therefore the prediction quality is improved. The combined approach...

This work presents a backpropagation neural network trained to reproduce the reaction yield of aryl fluorides by the halex technique. The work shows that a ten\|dimensional input space is able to reproduce reasonably the observed reaction yields by employing statistics in artificial neural system. By means of a number of multilayer feedforward (MLF) networks rather than one, the disadvantages caused by network randomness are limited greatly, and therefore the prediction quality is improved. The combined approach is suitable for relatively small training set, which often causes overfitting and leads to unreliable prediction results.

The technology of support vector machines is being used to solve problems of pattern recognition. Posteriori probability of samples is important in pattern recognition. But standard support vector machines do not provide posteriori probability. Discussed below are several questions based upon posteriori probability in the support vector machine: (1) decomposing the nonlinear optimal problem of a large training sample set into two nonlinear optimal problems of small training set; (2) designing the algorithm...

The technology of support vector machines is being used to solve problems of pattern recognition. Posteriori probability of samples is important in pattern recognition. But standard support vector machines do not provide posteriori probability. Discussed below are several questions based upon posteriori probability in the support vector machine: (1) decomposing the nonlinear optimal problem of a large training sample set into two nonlinear optimal problems of small training set; (2) designing the algorithm to revise the traditional optimal hyperplane, and analyzing the rationality of the algorithm; and (3) showing the results from testing on three image data sets effectively.