Faced with the broad prospect in research & application of cloud computing, the paper has studied and analyzed architecture and core principle of Hadoop platform, and its existing typical job scheduling algorithm. Furthermore, to resolve the pre-configuration issue of the algorithm, improved algorithm is proposed based on Naive-Bayes classification. According to the simulation test result, the improved algorithm is proven to be with satisfied performance, easy to learn. It may reduce administrator's workload, improve management efficiency, and reduce human error.