A New Algorithm on Lagged Correlation Analysis Between Time Series: TPFP

Lagged correlation analysis plays an important role in data mining based on time series, which can be used extensively in real life such as weather forcast, stock market analysis, network analysis, moving object tracking, sensor monitoring, and so on. Lagged correlation analysis is to find out the time series which are correlated with lags. Here three phenomena in lagged correlation analysis between time series, namely, continuous distribution of lags, lag mutation and mutation amplitude distribution feature, are found based on extensive experiments. It is proved that existing research work can achieve desirable performance when the lag is small, but there is sometimes large error when the lag is large. Furthermore, available methods can not deal with the occasion of lag mutation, which means that the lag changes suddenly at certain point on the lag correlation curve and is much different from before. This brings many difficulties for those existing methods. Based on the above three phenomena, three points forecast-based probing (TPFP) is proposed here to overcome the disadvantages of the existing methods, which is able to achieve small error when the lag is large, and also it can perform well on the occasion of lag mutation. Extensive experiments show that TPFP can achieve better performance than available methods.