In order to improve the analysis of reservoir properties and oil exploration and development process, this paper analyzes data and finds relationships between reservoir properties using Spark parallel computing framework and data mining algorithm, and classifies and predicts different reservoir segments. The main work in this paper includes: building the Spark distributed clustering and data processing and analysis platform, Spark being a popular big data parallel computing framework, which can achieve fast and accurate data mining tasks compared with some traditional analysis methods and tools; establishing a multidimensional outlier detection function according to the characteristics of reservoir data and adding a new discriminant attribute Pr; proposing a cross-recall training model and optimized cost function for logistic regression classification in dealing with the imbalanced data. KR-SMOTE is used to oversample for decession tree classification that both improve the classification precision.