Objective:

To perform the classification and mapping of body sway parameters from static posturography data to differentiate unilateral transtibial amputees from non-amputees using machine learning and data mining techniques.

Methodology:

Body sway was measured in 74 individuals, 37 landmine amputees and 37 healthy controls. Stability was classified by group using five machine learning algorithms. A continuous two-dimensional map of posture alterations was developed using Shannon's information theory, the U Mann-Whitney test (p <0,05) was used to identify differences between groups.

Results:

Five machine learning algorithms (decision tree, decision rules, neural network, vector support machine and clustering) were trained. Validation and comparison were carried out with the metrics obtained from the confusion matrix, using cross-validation to obtain two subsets. The most discriminatory posture condition was classified as displacement of the center of pressure (CoP) on the non-amputated side, anteroposterior direction. The algorithm with the highest performance was the vector support machine and the one with the lowest performance was the cluster; however, all the models performed group classification with an F1 score greater than 0,4.

Conclusions:

Mapping of sway displacement characteristics into 2D space revealed clear clusters between amputees and controls, confirming that machine learning can aid in the classification of clinical sway patterns measured with static posturography. Computational models allow to objectively evaluate the stability, as well as to recognize the contribution of the contralateral limb in the control of the static bipedal posture, since it compensates for the non-existence of the ipsilateral afferents and efferents.

Financing:

Scientific research article derived from the research "Characterization of Stability in Unilateral Transtibial Amputees", funded by "Francisco José de Caldas District University".