Predicting groundwater level fluctuations and estimating groundwater recharge are necessary for and effective management of groundwater resources. Applications of the water table fluctuation(WTF) method to groundwater recharge estimation are limited when time series data of groundwater level is discontinuous of abnormal in the present study, we designed a method to correct abnormal data using time series models for groundwater recharge estimation. An artificial neural network and a support vector machine were employed for time series model development and the hybrid water table fluctuation method (h-WTF) considering transient fillable porosity was utilized for groundwater recharge estimation. A comparison wtudy was conducted between three different techniques for groundwater recharge estimation: the classic WTF, h-WTF with observed data (h-WTF1), and h-WTF with corrected data (h-WTF2), using daily rainfall and groundwater level data of 5 groundwater monitoring stations in South Korea. Correlation coefficient values of observed and predicted groundwater level were as high as more than 0.8 for all the 5 stations. The result of the comparison study shows that the estimated ratio of recharge to rainfall ranges from 14.9 to 38.3 % for h-WTF1, and 21.8 to 50.5 % for h-WTF2 method. The estimated recharge ratios of h-WTF1 are smaller than h-WTF2. The reason is thought to be that the effect of exogenous factors to groundwater recharge except rainfall was filtered out through the time series model in h-WTF2 method.