Classifying walking pattern on different surfaces by optimising features extracted through IMU sensor data using SSA optimisation

It is difficult for patients with lower limb amputees to adjust while walking on diverse surfaces as they are fitted with prosthetic feet. It increases their chances of falling and can even result in further injuries. Various researchers are working on developing intelligent feet that can adapt to diverse surfaces. To implement this strategy, it is required to classify walking patterns based on different surfaces. In this study, nine different walking surfaces are selected for classification. Different statistical features and convolution features are extracted from vibration and gyroscopic signals acquired through the IMU sensor attached to the person’s body to select the best feature when they are walking on these surfaces. Afterwards, three shallow classifiers, namely SVM, ANN and LightGBM, are trained on these features for classifying these walking patterns on different surfaces. The LightGBM outperforms other shallow classifiers with a maximum accuracy 99.47% for convolutional features. Furthermore, the salp swarm algorithm (SSA) effectively reduces the feature dimension up to 49–56% maintaining the accuracy greater than 99%. Thus, the proposed data-driven methodology effectively classifies walking surfaces using the IMU sensor’s data.