Fig 1.
Flow-chart of the development of the machine learning algorithm.
Fig 2.
Flow-chart of the iteration process used to optimise the machine learning algorithm.
Fig 3.
Histogram indicating the error rates of discrimination for each individual iteration.
An iteration consisted of a different combination of 10 participants out of 20 for each the training and predicted database. The error is the percentage of variables that end up in the wrong category (shod or barefoot).
Fig 4.
Unoptimised result, showing data following discrimination undertaken on the entire collection of measurements, both for PCA and DFA, resulting in an error of 24%.
Fig 5.
DFA discrimination figure showing two bar charts where each bar is equivalent to a measured variable from a DF curve, integrated over all spectral frequencies.
Abbreviations are knee (KNE), ankle (ANK), angle (ANG), moment (MOM), power (POW), anterior-posterior (AP), medial-lateral (ML) and vertical (VERT).
Fig 6.
An illustrative representation of exemplar highly discriminating (A—sagittal plane ankle angle) and lower discriminating (B—sagittal plane knee angle) variables from a single participant during both shod (red limbs and lines) and barefoot (blue limbs and lines) running.
Dashed lines represent the instance in the gait cycle that the illustrations are taken from.
Fig 7.
Outcome of PCA following classification.
Each dot represents a trial of a participant and since each participant has conducted 10 trials (5 shod and 5 barefoot) and there was a total of 10 participants the figure illustrates the discrimination of 100 trials.
Fig 8.
Outcome of training database following discrimination, from the 10 participants with the smallest error in prediction.
Fig 9.
Outcome of discrimination for the 10 participants not used to generate the machine learning algorithm.
The scatter slightly greater than in Fig 3, but an excellent reliability in terms of correct discrimination.
Fig 10.
Combined display of trained and predicted data following discrimination.