Automated feature engineering
Feature engineering, for example, in automated machine learning, can include receiving streaming data representing at least one attribute detected by a sensor over time. Long term point statistics associated with the streaming data can be computed. The streaming data can be quantized into intervals of time windows and short term point statistics based on the intervals can be computed. The long term point statistics and the short term point statistics can be normalized. Dynamic time warping can be applied across the normalized long term point statistics and short term point statistics. A pair of probability distributions can be generated associated with the dynamic time warped normalized long term point statistics and short term point statistics. Based on distance between the mean values of the probability distributions, machine learning input features can be produced. The machine learning input features can be fed to train a machine learning model for detecting anomaly.
1 . A computer-implemented method comprising:
receiving streaming data representing at least one attribute detected by a sensor over time;
computing long term point statistics associated with the streaming data of a first time duration;
quantizing the streaming data into intervals of time windows, the intervals of time windows being of shorter time duration than the first time duration, and computing short term point statistics based on the intervals of time windows;
normalizing the long term point statistics and the short term point statistics;
applying dynamic time warping across the normalized long term point statistics and the short term point statistics;
generating a pair of probability distributions of the dynamic time warped normalized long term point statistics and the short term point statistics;
based on distance between mean values of the pair of probability distributions, producing machine learning input features, the input features comprising statistics representations processed from raw data, wherein the input features are learned in an unsupervised manner; and
feeding the machine learning input features to train a machine learning model for the machine learning model to detect anomalies in data streams before a failure event occurs to prevent the failure event,
wherein the machine learning model includes unsupervised anomaly detection model.
2 . The method of claim 1 , wherein the normalizing the long term point statistics and the short term point statistics includes determining percentage differences for the long term point statistics and the short term point statistics,
wherein the applying dynamic time warping includes applying dynamic time warping across the percentage differences determined for the long term point statistics and the short term point statistics, and
wherein the generating a pair of probability distributions includes generating a pair of probability distributions of the dynamic time warped percentage differences of the long term point statistics and the short term point statistics.
3 . The method of claim 1 , wherein the long term point statistics include at least one selected from the group consisting of mean, median, mode, and standard deviation.
4 . The method of claim 1 , wherein the short term point statistics includes at least one selected from the group consisting of mean, median, mode, and standard deviation.
5 . The method of claim 1 , wherein the machine learning model includes a model implementing an unsupervised machine learning algorithm.
6 . The method of claim 5 , wherein the unsupervised machine learning algorithm includes at least one selected from the group consisting of isolated forest, extended isolation forest, one class support vector machine (SVM), one class SVM (stochastic gradient descent (SGD)).
7 . The method of claim 1 , wherein the machine learning input features include at least one selected from the group consisting of fixed average, moving average, standard deviation for fixed average, standard deviation for moving average, z value for moving average, z value for fixed average, mode, median, percentage difference between median and moving average, warping path distance length.
8 . A system comprising:
at least one processor; and
at least one memory device coupled with the at least one processor;
the at least one processor configured to at least:
receive streaming data representing at least one attribute detected by a sensor over time;
compute long term point statistics associated with the streaming data of a first time duration;
quantize the streaming data into intervals of time windows, the intervals of time windows being of shorter time duration than the first time duration, and compute short term point statistics based on the intervals of time windows;
normalize the long term point statistics and the short term point statistics;
apply dynamic time warping across the normalized long term point statistics and the short term point statistics;
generate a pair of probability distributions of the dynamic time warped normalized long term point statistics and the short term point statistics;
based on distance between mean values of the pair of probability distributions, produce machine learning input features, the input features comprising statistics representations processed from raw data, wherein the input features are learned in an unsupervised manner; and
feed the machine learning input features to train a machine learning model for the machine learning model to detect anomaly in data streams before a failure event occurs to prevent the failure event,
wherein the machine learning model includes unsupervised anomaly detection model.
9 . The system of claim 8 , wherein the at least one processor configured to:
determine percentage differences for the long term point statistics and the short term point statistics, in normalizing the long term point statistics and the short term point statistics;
apply dynamic time warping across the percentage differences determined for the long term point statistics and the short term point statistics, in applying dynamic time warping; and
generate a pair of probability distributions of the dynamic time warped percentage differences of the long term point statistics and the short term point statistics, in generating a pair of probability distributions.
10 . The system of claim 8 , wherein the long term point statistics include at least one selected from the group consisting of mean, median, mode, and standard deviation.
11 . The system of claim 8 , wherein the short term point statistics includes at least one selected from the group consisting of mean, median, mode, and standard deviation.
12 . The system of claim 8 , wherein the machine learning model includes a model implementing an unsupervised machine learning algorithm.
13 . The system of claim 8 , wherein the machine learning input features include at least one selected from the group consisting of fixed average, moving average, standard deviation for fixed average, standard deviation for moving average, z value for moving average, z value for fixed average, mode, median, percentage difference between median and moving average, warping path distance length.
14 . A computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions readable by a device to cause the device to:
receive streaming data representing at least one attribute detected by a sensor over time;
compute long term point statistics associated with the streaming data of a first time duration;
quantize the streaming data into intervals of time windows, the intervals of time windows being of shorter time duration than the first time duration, and compute short term point statistics based on the intervals of time windows;
determine percentage differences for the long term point statistics and the short term point statistics;
apply dynamic time warping across the percentage differences determined for the long term point statistics and the short term point statistics;
generate a pair of probability distributions of the dynamic time warped percentage differences of the long term point statistics and the short term point statistics;
based on distance between mean values of the probability distributions, produce machine learning input features, the input features comprising statistics representations processed from raw data, wherein the input features are learned in an unsupervised manner; and
feed the machine learning input features to train a machine learning model for the machine learning model to detect anomaly in data streams before a failure event occurs to prevent the failure event,
wherein the machine learning model includes unsupervised anomaly detection model.
15 . The computer program product of claim 14 , wherein the device is caused to:
determine percentage differences for the long term point statistics and the short term point statistics, in normalizing the long term point statistics and the short term point statistics;
apply dynamic time warping across the percentage differences determined for the long term point statistics and the short term point statistics, in applying dynamic time warping; and
generate a pair of probability distributions of the dynamic time warped percentage differences of the long term point statistics and the short term point statistics, in generating a pair of probability distributions.
16 . The computer program product of claim 14 , wherein the long term point statistics include at least one selected from the group consisting of mean, median, mode, and standard deviation.
17 . The computer program product of claim 14 , wherein the short term point statistics includes at least one selected from the group consisting of mean, median, mode, and standard deviation.
18 . The computer program product of claim 14 , wherein the machine learning model includes a model implementing unsupervised machine learning algorithm.
19 . The computer program product of claim 14 , wherein the machine learning input features include at least one selected from the group consisting of fixed average, moving average, standard deviation for fixed average, standard deviation for moving average, z value for moving average, z value for fixed average, mode, median, percentage difference between median and moving average, warping path distance length.