How to realize privacy-preserving data mining becomes a research hotspot in a distributed environment. A model is proposed to realize privacy-preserving decision-tree classifying when data are vertically partitioned. In this model, a privacy-preserving decision-tree is proposed, which is composed of Global-Table stored by an obvious semi-honest partner and several local decision-trees stored by different sites. By using indexed array and private data comparison protocol, decision-tree generation and classification can be realized without uncovering the original information. Theoretical analysis and experimental results demonstrate the proposed model provides good capabilities of privacies preserving, accuracy and efficiency.

Aiming at the large computation of the maximum fuzzy entropy thresholding method, a fast algorithm for the maximum fuzzy entropy thresholding method is presented. It is based on the analysis of the character of the S-type function and the properties of the fuzzy entropy. The fast algorithm reduces the time complexity from O(L⁴) to O(L³). Meanwhile, the fast algorithm avoids the defaults of reaching the local extrema by the optimization methods. Therefore, the fast algorithm raises the speed and maintains the segmentation performance of the maximum fuzzy entropy thresholding method.

To improve the accuracy and the speed of on-line chaotic time series prediction via radial basis function (RBF) network, a sequential learning algorithm is presented for on-line constructing variable structure RBF network. A sliding window is constructed. By learning real-time updated data in the window, the parameters of the connecting weights, number of hidden units and center locations are dynamically tuned. The algorithm achieves parsimonious RBF network quickly, while only a small number of tuning parameters are employed. The variable structure network is applied to Mackey-Glass chaotic time series on-line prediction. The results demonstrate that network possesses satisfactory on-line dynamic identification and prediction performance.

The minimax probabilistic machine is a classifier based on minimizing the misclassification probability. The problem of 1-dimensional minimax probability machine is firstly discussed. Then a theory on minimax probabilistic image segmentation is presented. A method for developing criterion function for image thresholding is proposed. Meanwhile, the minimax probabilistic image thresholding algorithm is also proposed and it ensures the maximal lower bound for correctly classifying pixels. Experimental results show the effectiveness of the proposed algorithm.

The accuracy of least squares support vector machine(LSSVM) is influenced easily by gross errors and noises superimposed on value measurement of plant output when LSSVM is applied to the dynamic process online modeling directly. Aiming at that problem, robust online process modeling method using non-bias LSSVM is presented after the characteristics of sample sequence structure and noise action are analyzed. During the prediction period, abnormal measure data are recognized and recovered, and measure data containing noises are detected and rectified according to the relation between the predicting error and the set threshold value. Consequently, noises in samples are decreased,and online LSSVM tracks dynamics of process better. The modeling method is robust, and it decreases the effect of gross error and Gaussian white noise on the prediction accuracy of LSSVM to improve the prediction accuracy. The numerical simulation shows the validity and advantage of the proposed method.