Flexible Insole with Soft, Integrated, Polymeric Pressure Sensors to Discriminate Different Types of Gaits Using Machine Learning Techniques

Wearable sensors offer promising solutions for continuous gait monitoring in healthcare, rehabilitation, and sports. In this paper, a novel, soft, flexible, and easily made wearable sensor to detect the pressure generated by human gait was developed. The sensor consists of a soft polyurethane foam functionalized with a PEDOT:PSS aqueous solution designed to integrate the active material and develop a piezoresistive device. Eight soft sensors have been integrated into an insole, with dedicated readout electronic circuitry, including an accelerometer, a gyroscope, and a Bluetooth communication tag. Seven participants’ gaits were analyzed, and for each one, three types of walks were recorded: normal, prone, and supine. The data were analyzed with Machine Learning (ML) techniques, combining plantar pressure, accelerometer, and gyroscope features, to validate the device in discriminating among the three types of gaits. Ten supervised classifiers were assessed, such as Random Forest (RF), Support Vector Machine (SVM), Multi-layer Perceptron (MLP), K-Nearest Neighbors, and Gradient Boosting, using a sliding-window feature extraction pipeline and a chronological, per-recording train-test split designed to avoid information leakage between overlapping windows. Hyperparameter optimization was carried out via Randomized Search with a stratified 5-fold cross-validation protocol. The device has turned out to be suitable for detecting the three types of gaits thanks to the good results obtained by leveraging ML algorithms, reaching a classification accuracy of 90.38% on held-out data in a user-specific (subject-dependent) setting. In a stricter Leave-One-Subject-Out evaluation on participants never seen during training, accuracy reached 67.66%, about twice the chance level, without any labeled data from the new user. These results demonstrate the proposed flexible pressure system’s capability for continuous gait monitoring and support its use in healthcare diagnostics, rehabilitation evaluation, and sports biomechanics.