Beta This is a beta feature, which means we’re still developing it. Some functionality might change.
Data Analyzer allows you to create machine learning models that uncover patterns in your data to predict outcomes. Each model you create is added to the ML Models table in the Data Analyzer app. View the results from a model by clicking the eye icon for the record in the ML Models table. The model record shows you insights on the outcome you created the model to predict, including statistics on the the results and information on the model itself.
This article explains the data that is returned by the model, and how to understand the results:
General
The General section provides information on:
- The metadata on the model—Including the name you entered for the model, its ID number, and the timestamp of when the model was run.
- The data that was analyzed—Including the table analyzed, the outcome field and value you selected, and the percent of records that contained the outcome field and value.
- The performance of the model—Including the model performance rating and the model performance score. These measurements indicate the probability that the model will predict the correct outcome in two random records with opposite outcomes. A higher score indicates the model is better at predicting outcomes correctly.
Factor insights
The Factor insights tab shows you which factors impact the likelihood of the outcome field and value appearing in your data. These factors are statistical correlations between the outcome and field-value pairs in records that contain the outcome.
Insights follow a format of: When "field name" is "value X", your outcome is
"more or less" likely to happen.
You can also view the correlation and the confidence for each factor:
- Correlation—Indicates the strength of the relationship between the factor and the outcome. High correlation does not imply causation.
- Confidence—Indicates how likely the results did not occur by chance alone.
View statistical measures
Click the eye icon to view statistical measures for each factor listed in the table.
You'll find information on:
- Outcome rate—The percent of records that contain the outcome when the factor is present. You can also view the percent of records that contain the outcome across all records, and the percent of records that contain the outcome when the factor is not present.
- % of model weight contributed by this factor—How much of the model depends on the factor. You can find more information on the model weight in the Model details tab.
- pVal—The likelihood this factor is the same between records with and without the outcome value.
-
Lift—The outcome rate with the factor present compared to the average outcome rate for all records. This value is calculated with the formula:
([Outcome Rate with Factor]-[Outcome Rate for All Records])/[Outcome Rate for All Records]
Model details
The Model details tab shows you which factors have a high probability of impacting the outcome field and value. You can use the chart to view the model weight by field, or review the table to understand the model weight by factor. Only statistically significant factors are shown on this tab, which means that not all of the insights from the Factors insights tab are listed here.
Example: Say we are analyzing the outcome Attrition to understand why employees are leaving the company. In the bar chart below, the fields Regular Overtime, Years of Service, and Age have the highest probability of influencing attrition, according to historical data. You can drill into the data by clicking on a factor in the table to understand the impact on the outcome.
In this example, when Years of Service is 2 years or less, attrition is likely to be higher. When Years of Service is 5-9 years, attrition is likely to be lower.
Model deployment
The Model deployment tab contains the prediction formula used to calculate the probability of the outcome occurring based on trends in your data. The formula contains a representation of the machine learning model which is based on logistic regression.
If you decide to deploy this model in your table, you can use the formula to show a percent prediction for the outcome on each record in the data source.
Other factors analyzed
The Other factors analyzed section shows you other field and value combinations in the data source that may impact your outcome, but do not have at least 80% confidence.