Gait is a significant factor that affects human health, and monitoring a person's gait with sensing devices during daily life can detect abnormal gait events that affect numerous physical health problems. In particular, flat feet can cause changes in alignment conditions of the foot, ankle, leg, pelvis and spine. The primary problem with previous studies of wearable devices for measuring gait have focused on quantitatively monitoring the degree of gait rather than the limited gait ability. The existing method of feeding back the degree of gait or activity does not consider the severity of the subject and is insufficient for qualitative evaluation or training of gait. The significance of this study is development of convenient detecting and long-term tracking tools that can be used by both patients and clinicians for prescreening flat feet and monitoring the progress of flat feet treatment. For wearable devices for flatfoot detection to be most effective, detection systems and algorithms must be accurate, robust, reliable and computationally-efficient. In this paper, we developed an integrated smart wearable gait-monitoring device comprised of three sensors: front force, rear force, and an ankle flex sensor. We propose a new flat feet detection methodology based on a dynamic sensing window and a deep neural network with scaled principal component analysis (PCA). We tested 24 subjects, including both those with healthy gait and flat-feet-affected gait. Our study shows that the proposed sensing devices could be worn comfortably. The proposed deep neural network (DNN) model outperformed the other five classifier algorithms considered, and the area under the curve (AUC) value of the method was 87.1%. This wearable device can thus be easily and simply used both by patients and doctors to monitor the progress of flat feet and prescreen for possible gait problems in daily life.