Optimizing performance of event detection by sensor data analytics

• US 10,504,036 B2
• Filed: 01/06/2016
• Issued: 12/10/2019
• Est. Priority Date: 01/06/2016
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Limonad, Lior, Mashkif, Nir, Wasserkrug, Segev E, Zadorojniy, Alexander, Zeltyn, Sergey
• Primary Examiner(s): Waldron, Scott A.
• Assistant Examiner(s): Garner, Casey R.
• Application Number:
  US14/988,762
• Publication Number:
  US 20170193395A1
• Time in Patent Office: 1,434 Days
• Field of Search: None
• CPC Class Codes: G06N 20/00 Machine learning

Abstract

A computer-implemented method, computerized apparatus and computer program product, the method comprising: obtaining data measured by one or more sensors; segmenting the data into a plurality of sliding windows; extracting one or more features from each of the plurality of sliding windows; analyzing, by a machine learning process, the extracted features to determine, for each sliding window, an activity detection in the sliding window; and determining an activity detection result in the data to be positive responsive to activity detection by the machine learning process in at least a number M of sliding windows out of a number N of consecutive sliding windows, wherein M>1.

Claims

1. A computer-implemented method comprising:

obtaining data measured by one or more sensors, the data corresponding to a blood glucose level;

segmenting the data into a plurality of sliding windows;

extracting one or more features from each of the plurality of sliding windows;

analyzing, by a machine learning process, the extracted features to determine, for each sliding window, an activity detection in the sliding window; and

determining an activity detection result in the data to be positive responsive to activity detection by the machine learning process in at least a number M of sliding windows out of a number N of consecutive sliding windows, wherein M>

1 and N>

0, wherein M and N are activity detection parameters;

automatically administering a drug to a patient in response to the activity detection result indicating the blood glucose level is outside of a predetermined range;

determining a number of false positives and a number of true positives encountered over a plurality of activity detection results; and

optimize the activity detection parameters by, at least in part, adjusting M and N to maximize the number of true positives while maintaining the number of false positives encountered over a specified time period below a threshold number of false positives.

2. The computer-implemented method of claim 1, further comprising:

receiving a time-based constraint on false positive rate of activity detections, wherein the time-based constraint comprises the threshold number of false positives allowed over the specified time window.

3. The computer-implemented method of claim 1, wherein the activity detection parameters further comprise at least one parameter selected from the group consisting of:

a sliding window size;

an overlap length between consecutive sliding windows;

a type of classification algorithm utilized by the machine learning process;

a parameter of classification algorithm utilized by the machine learning process; and

type of features extracted.

4. The computer-implemented method of claim 1, wherein said determining comprises the steps of:

(a) selecting a value for the activity detection parameters prior to said segmenting, extracting, analyzing, and determining;

(b) performing said segmenting, extracting, analyzing, and determining in accordance with the values to obtain a plurality of activity detection results; and

(e) repeating steps (b) to (d) with one or more different values.

5. The computer-implemented method of claim 1, wherein the data is supervised.

6. The computer-implemented method of claim 1, wherein said optimizing the activity detection parameters comprises:

using a first portion of the data as training set for the machine learning process; and

using a second portion of the data as testing set for the machine learning process.

7. The computer-implemented method of claim 1, wherein the one or more sensors are wearable sensors.

8. The computer-implemented method of claim 1, further comprising preprocessing the data prior to said segmenting, the preprocessing comprising at least one step from the group consisting of:

cleaning; and

multi-dimensional synchronization.

9. The computer-implemented method of claim 1, further comprising issuing an alert responsive to determining the activity detection result to be positive.

10. A computer-implemented method comprising:

obtaining data measured by one or more sensors, the data corresponding to a blood glucose level;

receiving a time-based constraint on false positive rate of activity detections, wherein the time-based constraint comprises a maximum number of false positives over a specified time window;

determining at least one parameter maximizing true positive rate of activity detections subject to the time-based constraint, wherein the at least one parameter comprises a number M of sliding windows and a number N of consecutive sliding windows, wherein M>

1 and N>

0, said determining comprises the steps of;

(a) selecting a value for the at least one parameter;

(b) segmenting the data into a plurality of N sliding windows;

(c) extracting one or more features from each of the plurality of N sliding windows;

(d) analyzing, by a machine learning process, the extracted features to determine, for each sliding window, an activity detection result in the sliding window to obtain a plurality of activity detection results, wherein a positive result is indicated if the plurality of activity detection results are positive for at least N consecutive windows;

(e) calculating a number of false positives for the plurality of activity detection results;

(f) calculating true positive rate of the plurality of activity detection results; and

(g) repeating steps (b) to (f) with one or more different values for M and N to identify values of M and N that maximize the true positive rate while maintaining the number of false positives encountered over the specified time period below the maximum number of false positives;

wherein said segmenting, extracting, and analyzing are performed in accordance with the values; and

automatically administering a drug to a patient in response to the activity detection result indicating the blood glucose level is outside of a predetermined range.

11. A computerized apparatus having a processor, the processor being adapted to perform the steps of:

obtaining data measured by one or more sensors, the data corresponding to a blood glucose level;

segmenting the data into a plurality of sliding windows;

extracting one or more features from each of the plurality of sliding windows;

analyzing, by a machine learning process, the extracted features to determine, for each sliding window, an activity detection in the sliding window;

determining an activity detection result in the data to be positive responsive to activity detection by the machine learning process in at least a number M of sliding windows out of a number N of consecutive sliding windows, wherein M>

1 and N>

0, wherein M and N are activity detection parameters; and

automatically administering a drug to a patient in response to the activity detection result indicating the blood glucose level is outside of a predetermined range;

determining a number of false positives and a number of true positives encountered over a plurality of activity detection results; and

optimize the activity detection parameters by, at least in part, adjusting M and N to maximize the number of true positives while maintaining the number of false positives encountered over a specified time period below a threshold number of false positives.

12. The computerized apparatus of claim 11, wherein the activity detection parameters further comprise at least one parameter selected from the group consisting of:

a sliding window size;

an overlap length between consecutive sliding windows;

a type of classification algorithm utilized by the machine learning process;

a parameter of classification algorithm utilized by the machine learning process; and

type of features extracted.

13. The computerized apparatus of claim 11, wherein said determining comprises the steps of:

(a) selecting a value for the activity detection parameters prior to said segmenting, extracting, analyzing, and determining;

(b) performing said segmenting, extracting, analyzing, and determining in accordance with the values to obtain a plurality of activity detection results; and

(e) repeating steps (b) to (d) with one or more different values.

14. A computerized apparatus having a processor, the processor being adapted to perform the steps of:

obtaining data measured by one or more sensors, the data corresponding to a blood glucose level;

receiving a time-based constraint on false positive rate of activity detections, wherein the time-based constraint comprises a maximum number of false positives over a specified time window;

determining at least one parameter maximizing true positive rate of activity detections subject to the time-based constraint, wherein the at least one parameter comprises a number M of sliding windows and a number N of consecutive sliding windows, wherein M>

1 and N>

0, said determining comprises the steps of;

(a) selecting a value for the at least one parameter;

(b) segmenting the data into a plurality of N sliding windows;

(c) extracting one or more features from each of the plurality of N sliding windows;

(d) analyzing, by a machine learning process, the extracted features to determine, for each sliding window, an activity detection result in the sliding window to obtain a plurality of activity detection results, wherein a positive result is indicated if the plurality of activity detection results are positive for at least N consecutive windows;

(e) calculating a number of false positives for the plurality of activity detection results;

(f) calculating true positive rate of the plurality of activity detection results; and

(g) repeating steps (b) to (f) with one or more different values for M and N to identify values of M and N that maximize the true positive rate while maintaining the number of false positives encountered over the specified time period below the maximum number of false positives;

wherein said segmenting, extracting, and analyzing are performed in accordance with the values; and

automatically administering a drug to a patient in response to the activity detection result indicating the blood glucose level is outside of a predetermined range.

15. A computer program product comprising a computer readable storage medium retaining program instructions, which program instructions when read by a processor, cause the processor to perform a method comprising:

obtaining data measured by one or more sensors, the data corresponding to a blood glucose level;

segmenting the data into a plurality of sliding windows;

extracting one or more features from each of the plurality of sliding windows;

analyzing, by a machine learning process, the extracted features to determine, for each sliding window, an activity detection in the sliding window;

determining an activity detection result in the data to be positive responsive to activity detection by the machine learning process in at least a number M of sliding windows out of a number N of consecutive sliding windows, wherein M>

1 and N>

0, wherein M and N are activity detection parameters; and

automatically administering a drug to a patient in response to the activity detection result indicating the blood glucose level is outside of a predetermined range;

determining a number of false positives and a number of true positives encountered over a plurality of activity detection results; and

optimize the activity detection parameters by adjusting M and N to maximize the number of true positives while maintaining the number of false positives encountered over a specified time period below a threshold number of false positives.

16. A computer program product comprising a computer readable storage medium retaining program instructions, which program instructions when read by a processor, cause the processor to perform a method comprising:

obtaining data measured by one or more sensors, the data corresponding to a blood glucose level;

receiving a time-based constraint on false positive rate of activity detections, wherein the time-based constraint comprises a maximum number of false positives over a specified time window;

determining at least one parameter maximizing true positive rate of activity detections subject to the time-based constraint, wherein the at least one parameter comprises a number M of sliding windows and a number N of consecutive sliding windows, wherein M>

1 and N>

0, said determining comprises the steps of;

(a) selecting a value for the at least one parameter;

(b) segmenting the data into a plurality of N sliding windows;

(c) extracting one or more features from each of the plurality of N sliding windows;

(d) analyzing, by a machine learning process, the extracted features to determine, for each sliding window, an activity detection result in the sliding window to obtain a plurality of activity detection results;

(e) calculating a number of false positives for the plurality of activity detection results;

(f) calculating true positive rate of the plurality of activity detection results; and

(g) repeating steps (b) to (f) with one or more different values for M and N to identify values of M and N that maximize the true positive rate while maintaining the number of false positives encountered over the specified time period below the maximum number of false positives;

wherein said segmenting, extracting, and analyzing are performed in accordance with the values; and

automatically administering a drug to a patient in response to the activity detection result indicating the blood glucose level is outside of a predetermined range.