Big Data Analytics-as-a-Service (BDAaaS) is an emerging cloud computing service designed to provide big data analytics. BDAaaS offers approximate query services to meet the requirements of fast response but without high accuracy by users. So, service response in Service Agreement Level (SLA) are more stringent and BDAaaS faces greater risk of penalty payout. In this paper, we proposed a task scheduling model that takes SLA into account for the problem of sequential dependencies of some tasks. And a greedy strategy was used to improve the genetic algorithm to solve the model efficiently. Simulation experimental results show that the method in this paper can obtain task scheduling schemes with lower cost and SLA violation rate under limited resources.