The purpose of this paper is to evaluate and improve the accuracy of indoor positioning using smartphone sensors based on Pedestrian Dead Reckoning (PDR) method. In some specific situations, such as fires or power outages that disable infrastructure-based positioning techniques, using PDR method based on smartphone sensors that perform positioning continuously is a good solution.This paper focuses on determination of the user’s movement type to evaluate effective components of indoor positioning method. First, movement samples are evaluated with the feature-vectors of data from sensors and three classification algorithms (Decision Trees (DT), Support Vector Machine (SVM), and K-Nearest Neighbor (K-NN)). From the perspective of feature-vectors, the proposed features significantly improve the performance of three classification algorithms compared to previous research features. From the perspective of classification algorithm also Support Vector Machine had best performance with %99.3 accuracy, while spending the most time.  In the second phase, step detection is performed for norm acceleration values based on the definition of the upper and lower threshold and the time threshold. The directional component is also obtained by combining accelerometers, magnetometer and gyroscope sensors. Localization tests were performed while the user holding the phone in front of him with two states (normal walking, running) in three paths of different geometry (squares, circles and rectangles). The final accuracy obtained from normal walking test for three paths of square, circular and rectangular shapes was %4.8, %3.6, and %2, respectively. The final accuracy of the running mode was also obtained for three paths of square, circular and rectangular shapes equal to %8.4, %5.7, and %4, respectively.