Controlling a device based on log and sensor data

1. A system comprising:

one or more processors; and

one or more non-transitory computer-readable media maintaining executable instructions, which, when executed by the one or more processors, program the one or more processors to perform operations comprising:

receiving sensor data for a patient and at least one of:

log data for the patient; or

historical log data and historical sensor data for a plurality of patients;

determining at least one event classified as a non-uniform event in at least one of the log data or the historical log data;

determining a feature vector based on the sensor data and the at least one non-uniform event;

determining, based on the feature vector, an analysis result related to the patient; and

sending at least one of:

a notification or determination to a computing device; or

a control signal to a device for controlling a device.

2. The system as recited in claim 1 , wherein the operation of determining at least one event classified as the non-uniform event in at least one of the log data or the historical log data comprises:

determining a plurality of events in the at least one of the log data or the historical log data;

classifying events of the plurality of events as either uniform events or non-uniform events to determine the at least one non-uniform event; and

excluding at least one uniform event when determining the feature vector.

3. The system as recited in claim 2 , wherein the operation of classifying events of the plurality of events as either uniform events or non-uniform events comprises:

determining a plurality of events of different event categories, each occurrence of each event category associated with a timing;

determining a standard deviation of the timing of each occurrence of each event for each event category; and

classifying events associated with a standard deviation less than a threshold standard deviation as uniform and events associated with a standard deviation greater than the threshold standard deviation as non-uniform.

4. The system as recited in claim 1 , wherein the operation of determining the feature vector based on the sensor data and the at least one non-uniform event comprises:

determining time segments in a time window based on the at least one non-uniform event;

for each time segment, determining features from the sensor data corresponding to the time segment;

aggregating the features for time segments corresponding to a same order of occurrence of non-uniform events; and

determining the feature vector by concatenating the aggregated features for the time segments.

5. The system as recited in claim 4 , the operations further comprising:

for each time segment, determining an N-gram label based on the order of occurrence of the non-uniform events; and

aggregating the features for the time segments corresponding to a same order of occurrence of non-uniform events by aggregating features corresponding to the same N-gram, wherein the features include at least one of a mean of the sensor data within each time segment, a median of the sensor data within each time segment, or a standard deviation of the sensor data within each time segment.

6. The system as recited in claim 1 , the operations further comprising:

training a statistical model using the historical sensor data and at least one non-uniform event from the historical log data;

executing the statistical model using the sensor data and at least one non-uniform event from the log data in a selected time window;

determining time segments based at least in part on a hidden state of the statistical model;

for each time segment, determining one or more features from the sensor data;

aggregating the features based on matching hidden states; and

concatenating the aggregated features to determine the feature vector.

7. The system as recited in claim 1 , the operations further comprising:

training a neural network using the historical sensor data and at least one non-uniform event from the historical log data;

executing the neural network using the sensor data in a selected time window to determine one or more features from a cell of the neural network based on the sensor data; and

concatenating the features to determine the feature vector.

8. A method comprising:

receiving, by a processor, sensor data for a target and at least one of:

log data for the target; or

historical log data and historical sensor data for a plurality of other targets, wherein the log data and the historical log data include information related to the target generated by a human;

determining at least one event classified as a non-uniform event in at least one of the log data or the historical log data;

determining combined features based on the sensor data and the at least one non-uniform event; and

determining, based on the combined features, an analysis result related to the target.

9. The method as recited in claim 8 , further comprising:

determining a feature vector as the combined features;

determining the analysis result from the combined features based at least in part on at least one of regression or classification applied to the feature vector; and

based on the analysis result, sending at least one of:

a notification or determination related to the target to a computing device; or

a control signal to a device associated with the target for controlling the device.

10. The method as recited in claim 8 , further comprising:

determining a plurality of events in the at least one of the log data or the historical log data;

classifying events of the plurality of events as either uniform events or non-uniform events to determine the at least one event classified as the non-uniform event; and

excluding at least one event classified as a uniform event when determining the features.

11. The method as recited in claim 8 , further comprising:

determining time segments in a time window based on occurrences of a plurality of non-uniform events;

for individual time segments, determining features from the sensor data corresponding to the time segment;

aggregating the features according to time segments that correspond to a same order of occurrence of non-uniform events; and

determining the combined features by concatenating the aggregated features for the time segments.

12. The method as recited in claim 11 , further comprising determining the features from the sensor data for the individual time segments by determining at least one of:

a mean of the sensor data for each time segment,

a median of the sensor data for each time segment, or

a standard deviation of the sensor data for each time segment.

13. The method as recited in claim 8 , further comprising:

training a hidden Markov model (HMM) using the historical sensor data and at least one non-uniform event from the historical log data;

executing the HMM using the sensor data and at least one non-uniform event from the log data in a selected time window;

determining time segments based at least in part on a hidden state of the HMM;

for each time segment, determining one or more features from the sensor data;

aggregating the features based on matching hidden states; and

concatenating the aggregated features to determine the combined features.

14. The method as recited in claim 8 , further comprising:

training a neural network using the historical sensor data and at least one non-uniform event from the historical log data;

executing the neural network using the sensor data in a selected time window to determine one or more features from a cell of the neural network based on the sensor data; and

concatenating the features to determine the combined features.

15. A system comprising:

one or more processors; and

one or more non-transitory computer-readable media maintaining executable instructions, which, when executed by the one or more processors, program the one or more processors to:

receive sensor data for a target and at least one of:

log data for the target; or

historical log data and historical sensor data for a plurality of other targets;

determine at least one event classified as a non-uniform event in at least one of the log data or the historical log data;

determine a feature vector based on the sensor data and the at least one non-uniform event; and

determine, based on the feature vector, an analysis result related to the target.

16. The system as recited in claim 15 , wherein, based on the analysis result, the one or more processors are further programmed to send a control signal to a device associated with the target for controlling the device in response to a condition indicated by the analysis result.

17. The system as recited in claim 15 , wherein the one or more processors are further programmed to determine at least one event classified as the non-uniform event by:

determining a plurality of events in the at least one of the log data or the historical log data; and

classifying events of the plurality of events as either uniform events or non-uniform events to determine the at least one non-uniform event.

18. The system as recited in claim 15 , wherein the one or more processors are further programmed to:

determine time segments in a time window based on the at least one non-uniform event;

for individual time segments, determine features from the sensor data corresponding to the time segment;

aggregate the features for time segments corresponding to a same order of occurrence of non-uniform events; and

determine the feature vector by concatenating the aggregated features for the time segments.

19. The system as recited in claim 15 , wherein the one or more processors are further programmed to:

train a statistical model using the historical sensor data and at least one non-uniform event from the historical log data;

execute the statistical model using the sensor data and at least one non-uniform event from the log data in a selected time window;

determine time segments based at least in part on a hidden state of the statistical model;

for each time segment, determine one or more features from the sensor data;

aggregate the features based on matching hidden states; and

concatenate the aggregated features to determine the feature vector.

20. The system as recited in claim 15 , wherein the one or more processors are further programmed to:

train a neural network using the historical sensor data and at least one non-uniform event from the historical log data;

execute the neural network using the sensor data in a selected time window to determine one or more features from a cell of the neural network based on the sensor data; and

concatenate the features to determine the feature vector.