Introduction: Elderly people sometimes experience fall accidents since some of them cannot recognize their own stability in walking. Thus, fall prevention systems that measure and inform unstable walking have been developed. However, many previous systems required multiple sensors. The purpose of this study was to develop and test a gait recognition system using only a single inertial sensor for daily fall prevention. Methods: The proposed method recognizes an unstable gait by machine learning with trunk inertial data (three-axis acceleration and angular velocity) on the lower back and body measurement values (height and weight). The proposed method was trained and tested by the North American Congress on Biomechanics (NACOB) multi-surface walking dataset published by Jlassi et al. The trunk inertial data and body measurement values of 134 people in the NACOB public dataset were used in this study. The proposed method recognized two gait patterns on flat (stable) and bumpy (unstable) roads. Machine learning was implemented using the k-nearest neighbor algorithm (k=1). Training and testing were conducted via 5-fold cross validation. The accuracy and confusion matrix of gait recognition were evaluated. Results: The results showed that the proposed method could recognize stable and unstable gait patterns with greater than 80% accuracy. This accuracy was comparable to previous gait recognition. Conclusions: The results indicate the possibility that the proposed method can be used for daily gait recognition systems using a single inertial sensor. Acknowledgements: This study was supported by JSPS KAKENHI (Grant Number: 25K16012).