Insight Mining Using Machine Learning

1 . A method, comprising:

selecting a metric predictor model for predicting values of a data metric, the metric predictor model trained using historical data related to the data metric;

obtaining a predicted value of the data metric using the metric predictor model;

obtaining a current value of the data metric using data other than the historical data;

determining that a difference between the predicted value and the current value meets a reporting criterion; and

in response to determining that the difference meets the reporting criterion, outputting a notification descriptive of the difference.

2 . The method of claim 1 , wherein the metric predictor model is trained using the historical data related to the data metric by steps comprising:

generating time series data of the data metric using the historical data; and

obtaining the metric predictor model using the time series data.

3 . The method of claim 1 , wherein the data metric is a first data metric, further comprising:

receiving a second data metric;

training at least a subset of one or more models using historical data related to the second data metric;

determining that none of the trained at least the subset of the one or more models provides an expected prediction of the second data metric; and

outputting a notification indicating that predicting the second data metric using future data related to the second data metric will not be performed.

4 . The method of claim 1 , wherein the data metric is a first data metric, further comprising:

receiving a second data metric, wherein the second data metric comprises a dimension;

determining that the dimension is invalid as a candidate contributor dimension; and

outputting a notification indicating that the second data metric cannot be predicted based determining that the dimension is invalid.

5 . The method of claim 4 , wherein the dimension is determined to be invalid based on a cardinality of the dimension.

6 . The method of claim 4 , wherein the dimension is determined to be invalid based on a skewness of data that include the dimension.

7 . The method of claim 1 , further comprising:

determining that at least one prediction value of the data metric obtained using the metric predictor model deviates from at least one corresponding current value of the data metric; and

re-training the metric predictor model using at least the data other than the historical data.

8 . The method of claim 1 , further comprising:

determining a number of notifications descriptive of differences to be output by the metric predictor model using training data.

9 . The method of claim 8 , further comprising:

receiving, from a user, an indication of whether to use the metric predictor model to predict the data metric.

10 . A device, comprising:

a memory; and

a processor, the processor configured to execute instructions stored in the memory to:

select a metric predictor model for predicting values of a data metric, the metric predictor model trained using historical data related to the data metric;

obtain a predicted value of the data metric using the metric predictor model;

obtain a current value of the data metric using data other than the historical data;

determine that a difference between the predicted value and the current value meets a reporting criterion; and

in response to determining that the difference meets the reporting criterion, output a notification descriptive of the difference.

11 . The device of claim 10 , wherein the metric predictor model is trained using the historical data related to the data metric by instructions comprising instructions to:

generate time series data of the data metric using the historical data; and

obtain the metric predictor model using the time series data.

12 . The device of claim 10 , wherein the data metric is a first data metric, the processor further configured to execute instructions to:

receive a second data metric;

train at least a subset of one or more models using historical data related to the second data metric;

determine that none of the trained at least the subset of the one or more models provides an expected prediction of the second data metric; and

output a notification indicating that predicting the second data metric using future data related to the second data metric will not be performed.

13 . The device of claim 10 , wherein the data metric is a first data metric, the processor further configured to execute instructions to:

receive a second data metric, wherein the second data metric comprises a dimension;

determine that the dimension is invalid as a candidate contributor dimension; and

output a notification indicating that the second data metric cannot be predicted based determining that the dimension is invalid.

14 . The device of claim 13 , wherein the dimension is determined to be invalid based on a cardinality of the dimension.

15 . The device of claim 13 , wherein the dimension is determined to be invalid based on a skewness of data that include the dimension.

16 . The device of claim 10 , wherein the processor is further configured to execute instructions to:

determine that at least one prediction value of the data metric obtained using the metric predictor model deviates from at least one corresponding current value of the data metric; and

re-train the metric predictor model using at least the data other than the historical data.

17 . The device of claim 10 , wherein the processor is further configured to execute instructions to:

determine a number of notifications descriptive of differences to be output by the metric predictor model using training data.

18 . The device of claim 17 , wherein the processor is further configured to execute instructions to:

receive, from a user, an indication of whether to use the metric predictor model to predict the data metric.

19 . A non-transitory computer readable medium storing instructions operable to cause one or more processors to perform operations comprising:

selecting a metric predictor model for predicting values of a data metric, the metric predictor model trained using historical data related to the data metric;

obtaining a predicted value of the data metric using the metric predictor model;

obtaining a current value of the data metric using data other than the historical data;

determining that a difference between the predicted value and the current value meets a reporting criterion; and

in response to determining that the difference meets the reporting criterion, outputting a notification descriptive of the difference.

20 . The non-transitory computer readable medium of claim 19 , wherein the data metric is a first data metric, the operations further comprise:

receiving a second data metric, wherein the second data metric comprises a dimension;

determining that the dimension is invalid as a candidate contributor dimension; and

outputting a notification indicating that the second data metric cannot be predicted based determining that the dimension is invalid.