Data science has gained great attentions in many fields over the last decade, in this thesis, I further explored the use of data science technique in oil industry and developed three data mining applications that could be useful for reservoir modeling and exploratory data analysis. A detailed illustration of data mining algorithms such as Support Vector Machines (SVM), Probabilistic Neural Network (PNN) and Ensemble Learning algorithm is incorporated in the thesis. The performance of the proposed workflows are tested on real field data including Barnett Shale play and Mississippi Limestone. The first two applications are for the Barnett Shale play. For the first application, I used Support Vector Machines for the prediction on lithotypes derives from core data, the prediction algorithm takes a set of well log curves as input and lithotype as output. The test results showed that we achieved 76% accuracy in the blind test well, which indicates that we can identify lithotypes in uncored wells with high accuracy. For the second application, I proposed a workflow that used Ensemble Learning and Probabilistic Neural Network to make prediction on Total Organic Content (TOC) using a different set of well log curves. The blind test results showed that the predicted TOC zones share a great similarity with the core-based TOC measurement in the lab. The last application is for the Mississippi Lime in north-central Anadarko shelf of Oklahoma. I introduced a new porosity modeling workflow which combines Sequential Gaussian Simulation and Support Vector Machines. The results showed that my proposed workflow allows better use of exploratory data and make a more accurate estimation of the porosity in the reservoir model.