This study suggests a method for walking pattern recognition using the support vector machine (SVM) in a real-world situation, and aims to improve the algorithm based on the principal component analysis (PCA). For the experimental device, a smartphone was used to assess signals from the accelerometer, gyroscope, and magnetometer sensors. The participants were required to walk repeatedly on a treadmill at 3 km/h and 5 km/h for 3 min each. The results indicate that the PCA-SVM algorithm turned out to be more reliable than the non-PCA algorithm. In addition, the methodology used in this study, showed a robust performance, regardless of the sampling rates of the experimental device, which were from 10 to 50 Hz. The results of this study are expected to help researchers investigate physical movement patterns.