Select a model.

Trained Model

Select a previously trained model.

Train a new SVM for Sleep Staging or to predict Sleep Disordered Breathing. While training is in progress, you may navigate away from the page or close your browser. Trained models can be saved for later use in the Analysis section of the platform. Select an existing model to continue training (a warm start, with or without new data), or configure and train a new model.

Toggle Sleep Staging/Sleep Disordered Breathing

Model configuration.

Configuration

Select between training a Support Vector Classifier (SVC), a binary classifier to predict whether the AHI of a child is at least 5 or to classify recording as Sleep or Wake, or a Support Vector Regressor (SVR) to produce a point estimate of the AHI. Select a Linear, Gaussian, or Polynomial kernel. The Linear kernel allows for substantially better model interpretability, while the Gaussian kernel is more flexible and may capture non-linearities at the expense of interpretability. Polynomial kernels offer a balance in complexity and the degree of non-linearity is configurable.

Model Information

Gamma

Coef0

Epsilon

Bias

Features

Support Vectors

Feature Selection

	Feature	Description
	Desats	Continuous drops in the O2 saturation of at least 3%.
	Desat Depth	Cumulative percentage drop in the O2 saturation during desaturations.
	Desat Duration	Cumulative duration of desaturations to resaturation.
	Desat Area	Cumulative area of desaturations to resaturation.
	Baseline	90-second averaged O2 saturation baseline that does not include desaturations.
	OSA Band Power Median	Spectral component that encompasses half the power within the power spectral density of the OSA Band (0.01-0.035Hz) as computed using a Chirp-Z Transform.
	OSA Band Power Mean	First statistical moment of the power spectral density of the OSA Band.
	OSA Band Power Variance	Second statistical moment of the power spectral density of the OSA Band
	OSA Band Power Skewness	Third statistical moment of the power spectral density of the OSA Band.
	OSA Band Power Kurtosis	Fourth statistical moment of power spectral density of the OSA Band.
	OSA Band Power Maximum	Spectral component within the power spectral density of the OSA Band with the highest power.
	OSA Band Power Max Ratio	Spectral component within the power spectral density of the OSA Band with the highest power, relative to the highest power observed in the complete spectrum.
	OSA Band Power Sum	Sum of spectral component within the power spectral density of the OSA Band.
	OSA Band Power Sum Ratio	Sum of spectral component within the power spectral density of the OSA Band, relative to the sum of the power observed in the complete spectrum.
	Sats Mean	First statistical moment of the O2 saturation.
	Sats Variance	Second statistical moment of the O2 saturation.
	Sats Skewness	Third statistical moment of the O2 saturation.
	Sats Kurtosis	Fourth statistical moment of the O2 saturation.
	Sats CTM1	Central Tendency Measure of the O2 saturation where the radius equals 1. A measure of variability in the O2 saturation.
	Sats CTM3	Central Tendency Measure of the O2 saturation where the radius equals 3. A measure of variability in the O2 saturation.
	Sats CTM6	Central Tendency Measure of the O2 saturation where the radius equals 6. A measure of variability in the O2 saturation.
	Sats Complexity	Lempel-Ziv Complexity of the O2 saturation. A measure of repeating sequences in the O2 saturation.
	Sats RMSSD	Root mean square of successive differences in the O2 saturation. RMSSD is commonly used as a measure of heart rate variability.
	Sats Entropy	Shannon entropy of the O2 saturation. A measure of regularity.
	Sats Sample Entropy	Sample entropy of the O2 saturation where the template length equals 2, and the tolerance equals the product of 0.2 and the standard deviation of the O2 saturation. A measure of regularity.
	Sats Power Median	Spectral component that encompasses half the power within the power spectral density of the O2 saturation as computed using a Chirp-Z Transform.
	Sats Power Mean	First statistical moment of the power spectral density of the O2 saturation.
	Sats Power Variance	Second statistical moment of the power spectral density of the O2 saturation.
	Sats Power Skewness	Third statistical moment of the power spectral density of the O2 saturation.
	Sats Power Kurtosis	Fourth statistical moment of the power spectral density of the O2 saturation..
	Sats Power Maximum	Spectral component within the power spectral density of the O2 saturation with the highest power.
	Sats Power Sum	Sum of spectral component within the power spectral density of the O2 saturation.
	Sats DFA3	Detrended fluctuation analysis of the O2 saturation where the window is of size 3. A measurement of self-affinity in the O2 saturation.
	Sats DFA7	Detrended fluctuation analysis of the O2 saturation where the window is of size 7. A measurement of self-affinity in the O2 saturation.
	Sats DFA Trend Minimum	Minimum trend observed during DFA analysis of the O2 saturation.
	Sats DFA Trend Mean	Mean trend in DFA analysis of the O2 saturation.
	Sats DFA Trend Maximum	Maximum trend in DFA analysis of the O2 saturation.
	Sats AutoCorr 1	First-order autocorrelation of the O2 saturation. Measures the correlation of the O2 saturation with a delayed copy of itself.
	Sats AutoCorr 2	Second-order autocorrelation of the O2 saturation. Measures the correlation of the O2 saturation with a delayed copy of itself.
	Sats AutoCorr 3	Third-order autocorrelation of the O2 saturation. Measures the correlation of the O2 saturation with a delayed copy of itself.
	Sats AutoCorr 4	Fourth-order autocorrelation of the O2 saturation. Measures the correlation of the O2 saturation with a delayed copy of itself.
	Pulse Rate Mean	First statistical moment of the pulse rate.
	Pulse Rate Variance	Second statistical moment of the pulse rate.
	Pulse Rate Skewness	Third statistical moment of the pulse rate.
	Pulse Rate Kurtosis	Fourth statistical moment of the pulse rate.
	Pulse Rate CTM1	Central Tendency Measure of the pulse rate where the radius equals 1. A measure of variability in the pulse rate.
	Pulse Rate CTM3	Central Tendency Measure of the pulse rate where the radius equals 3. A measure of variability in the pulse rate.
	Pulse Rate CTM6	Central Tendency Measure of the pulse rate where the radius equals 6. A measure of variability in the pulse rate.
	Pulse Rate Complexity	Lempel-Ziv Complexity of the pulse rate. A measure of repeating sequences in the pulse rate.
	Pulse Rate RMSSD	Root mean square of successive differences in the pulse rate. RMSSD is commonly used as a measure of heart rate variability.
	Pulse Rate Entropy	Shannon entropy of the pulse rate. A measure of regularity.
	Pulse Rate Sample Entropy	Sample entropy of the pulse rate where the template length equals 2, and the tolerance equals the product of 0.2 and the standard deviation of the O2 saturation. A measure of regularity.
	Pulse Rate Power Median	Spectral component that encompasses half the power within the power spectral density of the pulse rate as computed using a Chirp-Z Transform.
	Pulse Rate Power Mean	First statistical moment of the power spectral density of the pulse rate.
	Pulse Rate Power Variance	Second statistical moment of the power spectral density of the pulse rate.
	Pulse Rate Power Skewness	Third statistical moment of the power spectral density of the pulse rate.
	Pulse Rate Power Kurtosis	Fourth statistical moment of the power spectral density of the pulse rate.
	Pulse Rate Power Maximum	Spectral component within the power spectral density of the pulse rate with the highest power.
	Pulse Rate Power Sum	Sum of spectral component within the power spectral density of the pulse rate.
	Pulse Rate DFA3	Detrended fluctuation analysis of the pulse rate where the window is of size 3. A measurement of self-affinity in the pulse rate.
	Pulse Rate DFA7	Detrended fluctuation analysis of the pulse rate where the window is of size 7. A measurement of self-affinity in the pulse rate.
	Pulse Rate DFA Trend Minimum	Minimum trend observed during DFA analysis of the pulse rate.
	Pulse Rate DFA Trend Mean	Mean trend in DFA analysis of the pulse rate.
	Pulse Rate DFA Trend Maximum	Maximum trend in DFA analysis of the pulse rate.
	Pulse Rate AutoCorr 1	First-order autocorrelation of the pulse rate. Measures the correlation of the pulse rate with a delayed copy of itself.
	Pulse Rate AutoCorr 2	Second-order autocorrelation of the pulse rate. Measures the correlation of the pulse rate with a delayed copy of itself.
	Pulse Rate AutoCorr 3	Third-order autocorrelation of the pulse rate. Measures the correlation of the pulse rate with a delayed copy of itself.
	Pulse Rate AutoCorr 4	Fourth-order autocorrelation of the pulse rate. Measures the correlation of the pulse rate with a delayed copy of itself.
	VAR Sats (-1) -> Pulse Rate	Vector Autoregression beta-coefficient for the lagged O2 saturation (-1) as a predictor of pulse rate, using a multivariable ordinary least squares regression model.
	VAR Sats (-2) -> Pulse Rate	Vector Autoregression beta-coefficient for the lagged O2 saturation (-2) as a predictor of pulse rate, using a multivariable ordinary least squares regression model.
	VAR Sats (-3) -> Pulse Rate	Vector Autoregression beta-coefficient for the lagged O2 saturation (-3) as a predictor of pulse rate, using a multivariable ordinary least squares regression model.
	VAR Sats (-4) -> Pulse Rate	Vector Autoregression beta-coefficient for the lagged O2 saturation (-4) as a predictor of pulse rate, using a multivariable ordinary least squares regression model.
	VAR Pulse Rate (-1) -> Sats	Vector Autoregression beta-coefficient for the lagged pulse rate (-1) as a predictor of O2 saturation, using a multivariable ordinary least squares regression model.
	VAR Pulse Rate (-2) -> Sats	Vector Autoregression beta-coefficient for the lagged pulse rate (-2) as a predictor of O2 saturation, using a multivariable ordinary least squares regression model.
	VAR Pulse Rate (-3) -> Sats	Vector Autoregression beta-coefficient for the lagged pulse rate (-3) as a predictor of O2 saturation, using a multivariable ordinary least squares regression model.
	VAR Pulse Rate (-4) -> Sats	Vector Autoregression beta-coefficient for the lagged pulse rate (-4) as a predictor of O2 saturation, using a multivariable ordinary least squares regression model.
	Desat Depth (%/hr)	Cumulative percentage drop in the O2 saturation, during desaturations, from a 90-second averaged baseline.
	CT90 (samples/hr)	Cumulative time below 90% O2 saturation per hour.
	CT95 (samples/hr)	Cumulative time below 95% O2 saturation per hour.
	ODI3	Desaturations of at least 3% from a 90-second averaged baseline.
	D85/hr	Number of desaturations below 85% O2 saturation per hour.
	D90/hr	Number of desaturations below 90% O2 saturation per hour.
	Desats/hr (any 3%)	Total desaturations of at least 3% per hour.

Add additional training data.

Dataset

Additional training data may be optionally added to the default training set. If this is a warm start of a previously trained model, these data will be included in the retraining process.

Training dashboard.

Bayesian Optimisation

C: The regularisation parameter that balances minimising training error (high C) and model generalisability (low C).Gamma: Balances decision boundary complexity (high Gamma) and simplicity (low Gamma). Coef0: A constant offset that balances complexity (low Coef0) and simplicity (high Coef0).Epsilon: Width of the epsilon-insensitive tube. Balances precision (low Epsilon) and generalisability (high Epsilon).

SVM Training Progress

Training in progress. You may use the analysis functionality or close your browser.

Finished

Parameters: C=25.663, Gamma=0.01, Coef0=1.171, Epsilon=3.01

Bayesian Optimisation of hyperparameters across iterations.

Expected Improvement

Predicted Mean Absolute Error (AHI)

Bayesian Optimisation is used to search the hyperparameter ranges provided. This process attempts to balance exploration and exploitation to select the optimal values for the trained model. Gaussian Process Regression (GPR) and the Expected Improvement acquisition function are used. For each set of hyperparameters, SVMs are trained using Sequential Minimal Optimisation (SMO) with several heuristics to ensure timely model convergence (Platt, 1998; Fan et al., 2005). K-fold cross-validation is used to produce a validation error which is input into the GPR that underpins the Bayesian Optimisation. The most optimal hyperparameters, as determined by the optimisation process, are then used to train the final model. SVMs are calibrated using cross-validation data when training is complete so that predictions are accompanied by estimates of uncertainty. A trained Ridge Logistic Regression model is embedded within SVC to produce a probability point-estimate, and a Laplace distribution is computed for SVR to produce confidence intervals that accompany the AHI point-estimate. Continuing the training of a previously trained model (i.e., a warm start) may significantly reduce training time. This process may be leveraged to quickly search for more performant hyper-parameter combinations, or to quickly redeploy a model with updated training data. You may save the trained model when training is complete. Saved models are accessible through the analysis feature of the platform.

Online Retraining

Configuration

Model Information

Feature Selection

Bayesian Optimisation

SVM Training Progress