This paper analyzes the effect of penetration rate to the estimation error in mobile phone based traffic state estimation systems. More concretely, the error-tolerance is analyzed based upon the penetration rate of participating mobile phones. In addition, a hybrid model by which not only real-time data but also the historical data utilized under a suitable data mining technique is introduced. This work also introduces an effective method for dynamically creating the historical dataset which is especially adequate for the aforementioned data mining model. This approach improves not only the effectiveness, robustness and the accuracy but also the scalability of the system. The evaluation reveals that the estimation error is sensitive to the penetration rate while the existing work did not mention.