In this paper, we address the oscillations in the signals of both motion capture and inertial measurement sensors. This characteristic is often observed when the range of motion reaches or exceeds approximately 85 degrees. Elimination of oscillation in the filtered output signal is of significance as it means the filtered signal can be applied directly by applications such as visualization, motion tracking, clinical report, etc. This paper proposed a system model using feature selection and machine learning algorithm to automatically detect and post-filter the oscillation. Feature selection aims to derive the most impactful wearable sensor data gathered from the accelerometer, magnetometer and gyroscope, using well-known classifiers such as Logistic Regression, Support Vector Machine and Multilayer Perceptron. The features and classification method that are most suited for the detection model are selected. Experimental results show that we can on average achieve up to 76% accuracy and are able to filter out the fluctuation; the latter is more efficient than manually post-processing the data. The trained detection model has thus proven its effectiveness in eliminating the noisy fluctuations and its potential to be used in real-time.