Stratified sampling is a sampling method where population is divided into homogenous subgroups called strata and the right number of instances are sampled from each stratum. For further explanation visit here.
This sampling is important to ensure that sampled dataset is representative of the entire population. To realise this point, consider an example of predicting the party would would win the election. Suppose there are 2 different class of voters, rural and urban class with 30% and 70% proportions, respectively. Now pre-poll survey is done in sampled population by asking people which party will they vote for. If 90% of the people who took survey are urban and 10% of them are rural, we may extrapolate the urban class opinions and undermine the power of rural voters. This may happen because survey could have been taken in digital mode and hence, mostly urban class people would have taken the survey. To avoid this disproportion we need to make sure the surveyed people also have the sample proportion as in the entire population. And this process of sampling in order to keep the same proportion of each class as present in population is called stratified sampling.