R-square is the percent of variability explained by the model and 64% then is not explained by the model, explained by other factors and possibly due to chance. This is important because the higher the r-squared the better the data fits the model. So with a low r-squared the data isn't the best fit for the model.