Imputing time series data by regional-gradient-guided bootstrapping algorithm

The problem of missing 2-dimensional time series data is one of the main problems existing in several real scientific and engineering studies. In this paper, a new technique for imputing the incomplete time series data is proposed. The imputing process combines two major steps. The first step is to estimate the potential imputing boundary regions based on the intersection of the slopes of non-missing neighbors. Then, a new bootstrap algorithm is applied to estimate the value of missing data. The experimental results show that our new algorithms outperforms in both accuracy and time efficiency when compared with cubic interpolation, multiple imputation(MI) and varies window similarity measure(VWSM) algorithms under various missing rates from 10% to 70%.