Forecasting Of Wireless Network Traffic Based On Local Minimax Probability Machine

In recent years, with the continuous promotion of application of wireless networktechnology, the security problem has become one of the most critical issues encountered in thewireless LAN development. In addition to the access control technology which is realizedbased on key management and authentication, network traffic prediction and anomaly detectionhas become important means to solve the problem. At the same time, the network trafficanalysis and prediction play an important role in large scale network capacity planning,network equipment design, network resources management and user behavior regulation.Because the wireless network traffic has obvious instability, poor regularity characteristics, thetraditional forecasting methods are not applicable. The minimax Probability MachineRegression is a new intelligent algorithm for time series prediction recently, but it has defects inmodel constructing, so this paper propose a local minimax probability machine predictionalgorithm for wireless network traffic prediction. The main contribution of this article is:First, calculate the number of the nearest neighbor points using AICi criterion, propose touse a high-dimensional nearest neighbor points selection algorithm based on KD-Tree. Thealgorithm first constructs the KD-Tree using the sample points reconstructed, and then searchthe k-nearest target points in the KD-Tree using certain search algorithm.Second, an improved “Ki” strategy is proposed. The strategy fully takes the influence of“neighbor” which is made by the curve direction to into account, it uses the cosine betweeneach phase point in NNPS and the target point to eliminate pseudo neighbor points.Finally, this paper proved the effectiveness of the improved local predict model, it can bemore accurately, much timely forecast the wireless network traffic.