Idiopathic toe walking (ITW) is a gait abnormality in which children toe touch at initial contact and demonstrate limited or no heel contact throughout the gait cycle. Toe walking results in poor balance, increased risk of falling, and developmental delays among children. Identifying toe walking steps during walking can facilitate targeted intervention among children diagnosed with ITW. With recent advances in wearable sensing, communication technologies, and machine learning, new avenues of managing toe walking behavior among children are feasible. In this study, we investigate the capabilities of Machine Learning (ML) algorithms in identifying initial foot contact (heel strike versus toe strike) utilizing wearable body sensors. Thirty-six children (Age 9.4±2.8 years) diagnosed with ITW participated in this study. Six ML algorithms, consisting of Support Vector Machines (SVM), decision tree (DT), random forest (RF), K-nearest neighbors (KNN), Multi-layer Perceptron (MLP), and Gaussian process (GP), could successfully classify initial contact walking patterns among ITW. We found that a simple KNN algorithm resulted in the highest accuracy of 92.92% and F1-score of 93.20% to differentiate toe walking gait versus best heel strike when using all four body sensors. We also found that toe walking resulted in higher variability in the sacral vertical accelerations among children diagnosed with ITW. Accurate quantification of toe walking steps in clinical applications is critical for assessing rehabilitation progress and designing new interventions for children diagnosed with ITW.