This study compares the results of analyzing tsunami simulations that are based on two approaches of characterizing earthquake slips, i.e., uniform (simplistic) and heterogeneous (complex) distributions. The aim of this study is to compare how heterogeneous and uniform distributed data affect the classification of tsunami maximum near-shore tsunami amplitudes. Due to the lack of historical earthquake and tsunami data to train the forecasting model, 4000 stochastic tsunami simulations are employed. The focused location is Iwanuma, Japan, where ocean bottom sensors (OBS) S-net network has been deployed. Multiple linear regression combined with Akaike Information Criterion (AIC) is applied to the simulated off-shore wave amplitude data to fit the model. The estimated tsunami amplitude is classified into four levels of warning classes. The performance of the models is quantified by the accuracy of the confusion matrices and is compared with the base model that only uses earthquake information. The forecasting accuracy can be improved by 30% when the wave amplitude data are used as additional information. The heterogeneous slip-based model reaches a higher accuracy than the uniform-slip based model. The result of this study is particularly valuable for setting up an OBS-based system for monitoring the physical phenomena of tsunamis and choosing heterogeneous as a preferable slip distribution when tsunami events are simulated.