Off-body Sleep Analysis for Predicting Adverse Behavior in Individuals with Autism Spectrum Disorder

Abstract

Poor sleep quality in Autism Spectrum Disorder (ASD) individuals is linked to severe daytime behaviors. This study explores the relationship between a prior night's sleep structure and its predictive power for next-day behavior in ASD individuals. The motion was extracted using a low-cost near-infrared camera in a privacy-preserving way. Over two years, we recorded overnight data from 14 individuals, spanning over 2,000 nights, and tracked challenging daytime behaviors, including aggression, self-injury, and disruption. We developed an ensemble machine learning algorithm to predict next-day behavior in the morning and the afternoon. Our findings indicate that sleep quality is a more reliable predictor of morning behavior than afternoon behavior the next day. The proposed model attained an accuracy of 74% and a F1 score of 0.74 in target-sensitive tasks and 67% accuracy and 0.69 F1 score in target-insensitive tasks. For 7 of the 14, better-than-chance balanced accuracy was obtained (p-value<0.05), with 3 showing significant trends (p-value<0.1). These results suggest off-body, privacy-preserving sleep monitoring as a viable method for predicting next-day adverse behavior in ASD individuals, with the potential for behavioral intervention and enhanced care in social and learning settings.

Figure 1:

Diagram illustrating the workflow for collecting movement data using a Raspberry Pi 4 Model B equipped with a 5MP OV5647 infrared camera. The camera has a resolution of 2592 × 1944 and a frame rate of 60fps. Frames are downsampled to 5×4 superpixels (with 4 superpixels wide and 5 superpixels high), and the frame rate is resampled to 1.5Hz for analysis. Both Global Difference Sum(GDS) and Local Difference Sum(LDS) are calculated. These features are aggregated for each time frame to generate a multidimensional time series. The resulting data is then securely uploaded to a HIPAA-compliant cloud. In the days following data collection, the behaviorist logs resident labels during two sessions (morning and afternoon) for three distinct target adverse behaviors: aggression, disruptive behavior, and self-injurious behavior (SIB).

Figure 2:

Flowchart illustrating the two-stage methodological approach for sleep analysis. The first stage involves feature extraction from Global Difference Sum (GDS) and Local Difference Sum (LDS), resulting in a 105-dimensional feature vector for each observation. Five specific features are extracted, each chosen for its relevance in sleep analysis. The second stage employs an ensemble of seven machine learning algorithms to make predictive assessments based on the extracted features.

Figure 3:

Confusion matrices for target-sensitive and target-insensitive models predicting the presence of adverse behavior at different times of the day: (a) Morning using the target-sensitive model, (b) Afternoon using the target-sensitive model, (c) Morning using the target-insensitive model, and (d) Afternoon using the target-insensitive model.

Figure 4:

Confusion matrices for predicting different adverse behaviors in the morning time. Confusion matrix for predicting (a) Aggression, (b) Self-injurious behavior, and (c) Disruptive behavior.

Figure 5:

Performance analysis of predictive models using different lengths of historical nights’ data(a) F1 Score and (b) Accuracy across different residents using various lengths of consecutive nights for prediction. The graphs showcase the performance outcomes, represented by F1 score and accuracy, for different residents when using varying lengths of consecutive nights’ data for prediction. It is notable that utilizing data from a 7-day period yields the highest results for both metrics, even slightly surpassing the performance achieved when only data from a single preceding night is used.