Exploring Random Forest Machine Learning for Fetal Movement Detection using Abdominal Acceleration and Angular Rate Data.

Fetal movement is a commonly monitored indicator of fetal wellbeing with reductions in fetal movement being associated with poor perinatal outcomes. However, more informative datasets of fetal movement are required for improved clinical decision making. Wearable sensors coupled with machine learning (ML) methods could support accurate detection of fetal movement. While prior work has demonstrated the feasibility of accelerometer-based detection, using angular rate data to train ML models has not been fully explored. The goal of this study was to train and validate ML models using acceleration and angular rate features for detection of fetal movement. Ten pregnant participants wore an array of four abdominal inertial measurement units (IMUs) and one chest reference, while holding a toggle for maternal perception of fetal movement. Three random forest classifiers were trained on acceleration features, angular rate features, and a combination of both feature sets, respectively. Classification performance was reported in terms of area under the receiver operating characteristic curve (AUROC) and standard performance metrics. Results showed that all three models achieved good performance (AUROC = 0.70-0.77). The model combining acceleration and angular rate features achieved a notably higher positive predictive value (PPV) compared to the other models developed, indicating discriminative power over either feature set alone.