Sleep Disordered Breathing

Analysis of nocturnal paediatric oximetry uses machine learning (ML).

If classification is selected, a binary classifier produces a probabilistic or non-probabilistic output. For non-probabilistic models such as Support Vector Classifiers (SVC), the calibrated probability of an AHI ≥5 is calculated using a Ridge Logistic Regressor trained using cross-validation data. It is embedded within the SVC and invoked during prediction to provide a calibrated probability.

If regression is selected, a trained model produces a point estimate of the AHI. Measures of uncertainty are provided using a Laplace distribution that is computed and embedded within the model shortly after training. It is invoked during prediction to provide a conservative confidence interval.

Model Explanations

Feature importance values or explanations attempt to describe the impact of individual features on the model output. Care should be taken when interpreting these values, and consideration should be given to the real world directionality of each feature when interpreting the sign of an explanation value.

To estimate the overall importance of each feature to a prediction, Global Explanations are approximated using the KernelSHAP method. The coefficients of a weighted linear model provide an approximation of how changes in each feature influences the model prediction.

To understand model behaviour near a specific prediction, Local Interpretable Model-agnostic Explanations (LIME) and a K-LASSO approach are used. Lasso regression is applied to select K (where K equals 5) features that are most relevant locally. A linear model is trained on perturbed samples around the original data, weighted by their proximity. The coefficients of this linear model form Local Explanations, providing a sparse estimate of how the prediction changes with small, localised variations in the selected features.