Create and Use Oracle Analytics Predictive Models

Oracle Analytics predictive models use several embedded machine learning algorithms to mine your data sets, predict a target value, or identify classes of records. Use the data flow editor to create, train, and apply predictive models to your data.

Topics:

What Are Oracle Analytics Predictive Models?

An Oracle Analytics predictive model applies a specific algorithm to a data set to predict values, predict classes, or to identify groups in the data.

You can also use Oracle machine learning models to predict data. See How Can I Use Oracle Machine Learning Models in Oracle Analytics?.

Oracle Analytics includes algorithms to help you train predictive models for various purposes. Examples of algorithms are classification and regression trees (CART), logistic regression, and k-means.

You use the data flow editor to first train a model on a training data set. After the predictive model has been trained, you apply it to the data sets that you want to predict.

You can make a trained model available to other users who can apply it against their data to predict values. In some cases, certain users train models, and other users apply the models.

Note:

If you're not sure what to look for in your data, you can start by using Explain, which uses machine learning to identify trends and patterns. Then you can use the data flow editor to create and train predictive models to drill into the trends and patterns that Explain found. See What is Explain?

You use the data flow editor to train a model:

First, you create a data flow and add the data set that you want to use to train the model. This training data set contains the data that you want to predict (for example, a value like sales or age, or a variable like credit risk bucket).
If needed, you can use the data flow editor to edit the data set by adding columns, selecting columns, joining, and so on.
After you've confirmed that the data is what you want to train the model on, you add a training step to the data flow and choose a classification (binary or multi), regression, or cluster algorithm to train a model. Then name the resulting model, save the data flow, and run it to train and create the model.
Examine the properties in the machine learning objects to determine the quality of the model. If needed, you can iterate the training process until the model reaches the quality you want.

Use the finished model to score unknown, or unlabeled, data to generate a data set within a data flow or to add a prediction visualization to a project.

Example

Suppose you want to create and train a multi-classification model to predict which patients have a high risk of developing heart disease.

Supply a training data set containing attributes on individual patients like age, gender, and if they've ever experienced chest pain, and metrics like blood pressure, fasting blood sugar, cholesterol, and maximum heart rate. The training data set also contains a column named "Likelihood" that is assigned one of the following values: absent, less likely, likely, highly likely, or present.
Choose the CART (Decision Tree) algorithm because it ignores redundant columns that don't add value for prediction, and identifies and uses only the columns that are helpful to predict the target. When you add the algorithm to the data flow, you choose the Likelihood column to train the model. The algorithm uses machine learning to choose the driver columns that it needs to perform and output predictions and related data sets.
Inspect the results and fine tune the training model, and then apply the model to a larger data set to predict which patients have a high probability of having or developing heart disease.

How Do I Choose a Predictive Model Algorithm?

Oracle Analytics provides algorithms for any of your machine learning modeling needs: numeric prediction, multi-classifier, binary classifier, and clustering.

Oracle's machine learning functionality is for advanced data analysts who have an idea of what they're looking for in their data, are familiar with the practice of predictive analytics, and understand the differences between algorithms.

Normally users want to create multiple prediction models, compare them, and choose the one that's most likely to give results that satisfy their criteria and requirements. These criteria can vary. For example, sometimes users choose models that have better overall accuracy, sometimes users choose models that have the least type I (false positive) and type II (false negative) errors, and sometimes users choose models that return results faster and with an acceptable level of accuracy even if the results aren't ideal.

Oracle Analytics contains multiple machine learning algorithms for each kind of prediction or classification. With these algorithms, users can create more than one model, or use different fine-tuned parameters, or use different input training datasets and then choose the best model. The user can choose the best model by comparing and weighing models against their own criteria. To determine the best model, users can apply the model and visualize results of the calculations to determine accuracy, or they can open and explore the related data sets that Oracle Analytics used the model to output. See What Are a Predictive Model's Related Data Sets?

Consult this table to learn about the provided algorithms:

Name	Type	Category	Function	Description
CART	Classification Regression	Binary Classifier Multi-Classifier Numerical	-	Uses decision trees to predict both discrete and continuous values. Use with large data sets.
Elastic Net Linear Regression	Regression	Numerical	ElasticNet	Advanced regression model. Provides additional information (regularization), performs variable selection, and performs linear combinations. Penalties of Lasso and Ridge regression methods. Use with a large number of attributes to avoid collinearity (where multiple attributes are perfectly correlated) and overfitting.
Hierarchical	Clustering	Clustering	AgglomerativeClustering	Builds a hierarchy of clustering using either bottom-up (each observation is its own cluster and then merged) or top down (all observations start as one cluster) and distance metrics. Use when the data set isn't large and the number of clusters isn't known beforehand.
K-Means	Clustering	Clustering	k-means	Iteratively partitions records into k clusters where each observation belongs to the cluster with the nearest mean. Use for clustering metric columns and with a set expectation of number of clusters needed. Works well with large datasets. Result are different with each run.
Linear Regression	Regression	Numerical	Ordinary Least Squares Ridge Lasso	Linear approach for a modeling relationship between target variable and other attributes in the data set. Use to predict numeric values when the attributes aren't perfectly correlated.
Logistic Regression	Regression	Binary Classifier	LogisticRegressionCV	Use to predict the value of a categorically dependent variable. The dependent variable is a binary variable that contains data coded to 1 or 0.
Naive Bayes	Classification	Binary Classifier Multi-Classifier	GaussianNB	Probabilistic classification based on Bayes' theorem that assumes no dependence between features. Use when there are a high number of input dimensions.
Neural Network	Classification	Binary Classifier Multi-Classifier	MLPClassifier	Iterative classification algorithm that learns by comparing its classification result with the actual value and returns it to the network to modify the algorithm for further iterations. Use for text analysis.
Random Forest	Classification	Binary Classifier Multi-Classifier Numerical	-	An ensemble learning method that constructs multiple decision trees and outputs the value that collectively represents all the decision trees. Use to predict numeric and categorical variables.
SVM	Classification	Binary Classifier Multi-Classifier	LinearSVC, SVC	Classifies records by mapping them in space and constructing hyperplanes that can be used for classification. New records (scoring data) are mapped into the space and are predicted to belong to a category, which is based on the side of the hyperplane where they fall.

Typical Workflow to Create and Use Oracle Analytics Predictive Models

Here are the common tasks for creating predictive models, and how to apply the models to data sets and use them in projects.

Task	Description	More Information
Train a model using sample data	Use one of the supplied algorithms to train a model to predict trends and patterns in your sample data.	Create and Train a Predictive Model
Evaluate a model	Use related data sets to evaluate the effectiveness of your model, and iteratively refine the model until you're satisfied with it.	Inspect a Predictive Model
Apply a model to your data using a data flow	Apply a training predictive model to your data to generate a data set that includes the predicted trends and patterns.	Apply a Predictive or Registered Oracle Machine Learning Model to a Data Set
Apply a predictive model to your project data	Use a scenario to add a predictive model to your project.	Add a Predictive Model to a Project

Create and Train a Predictive Model

Based on the problem that needs to be solved, an advanced data analyst chooses an appropriate algorithm to train a predictive model and then evaluates the model's results.

Arriving at an accurate model is an iterative process and an advanced data analyst can try different models, compare their results, and fine tune parameters based on trial and error. A data analyst can use the finalized, accurate predictive model to predict trends in other data sets, or add the model to projects.

Oracle Analytics provides algorithms for numeric prediction, multi-classification, binary-classification and clustering. For information about how to choose an algorithm, see How Do I Choose a Predictive Model Algorithm?

The algorithms aren't available until you install Oracle machine learning into your local Oracle Analytics Desktop directory. See How do I install Machine Learning for Desktop?

On the Home page, click Create, and then select Data Flow.
Select the data set that you want to use to train the model. Click Add.
Typically you'll select a data set that was prepared specifically for training the model and contains a sample of the data that you want to predict. The accuracy of a model depends on how representative the training data is.
In the data flow editor, click Add a step (+).
After adding a data set, you can either use all columns in the data set to build the model or select only the relevant columns. Choosing the relevant columns requires an understanding of the data set. Ignore columns that you know won't influence the outcome behavior or that contain redundant information. You can choose only relevant columns by adding the Select Columns step. If you're not sure about the relevant columns, then use all columns.
Navigate to the bottom of the list and click the train model type that you want to apply to the data set.
Select an algorithm and click OK.
If you're working with a supervised model like prediction or classification, then click Target and select the column that you're trying to predict. For example, if you're creating a model to predict a person's income, then select the Income column.
If you're working with an unsupervised model like clustering, then no target column is required.
Change the default settings for your model to fine tune and improve the accuracy of the predicted outcome. The model you're working with determines these settings.
Click the Save Model step and provide a name and description. This will be the name of the generated predictive model.
Click Save, enter a name and description of the data flow, and click OK to save the data flow.
Click Run Data Flow to create the predictive model based on the input data set and model settings that you provided.

Inspect a Predictive Model

After you create the predictive model and run the data flow, you can review information about the model to determine its accuracy. Use this information to iteratively adjust the model settings to improve its accuracy and predict better results.

Topics:

View a Predictive Model's Details

A predictive model's detail information helps you understand the model and determine if it's suitable for predicting your data. Model details include its model class, algorithm, input columns, and output columns

On the Home page, click Navigator, and then click Machine Learning.
Click the Models tab.
Click the menu icon for a training model and select Inspect.
Click the Details to view the model's information.

Assess a Predictive Model's Quality

View information that helps you understand the quality of a predictive model. For example, you can review accuracy metrics like model accuracy, precision, recall, F1 value, and false positive rate.

Oracle Analytics provides similar metrics irrespective of the algorithm used to create the model, thereby making comparison between different models easy. During the model creation process, the input data set is split into two parts to train and test the model based on the Train Partition Percent parameter. The model uses the test portion of the data set to test the accuracy of the model that is built.

Based on your findings in the Quality tab, you may need to adjust the model parameters and retrain it. See Create and Train a Predictive Model.

On the Home page, click Navigator, and then click Machine Learning.
Click the Models tab.
Click the menu icon for a training model and select Inspect.
Click the Quality tab to review the model's quality metrics.

What Are a Predictive Model's Related Data Sets?

When you run the data flow to create the Oracle Analytics predictive model's training model, Oracle Analytics creates a set of related data sets. You can open and create projects on these data sets to learn about the accuracy of the model.

See Find a Predictive Model's Related Data Sets.

Depending on the algorithm you chose for your model, related data sets contain details about the model such as prediction rules, accuracy metrics, confusion matrix, and key drivers for prediction. You can use this information to fine tune the model to get better results, and you can use related data sets to compare models and decide which model is more accurate.

For example, you can open a Drivers data set to discover which columns have a strong positive or negative influence on the model. By examining those columns, you find that some columns aren't treated as model variables because they aren't realistic inputs or that they're too granular for the forecast. You use the data flow editor to open the model and based on the information you discovered, you remove the irrelevant or too-granular columns, and regenerate the model. You check the Quality and Results tab and verify if the model accuracy is improved. You continue this process until you're satisfied with the model's accuracy and it's ready to score a new data set.

Different algorithms generate similar related data sets. Individual parameters and column names may change in the data set depending on the type of algorithm, but the functionality of the data set stays the same. For example, the column names in a statistics data set may change from Linear Regression to Logistic Regression, but the statistics data set contains accuracy metrics of the model.

These are the related data sets:

CARTree

This data set is a tabular representation of CART (Decision Tree), computed to predict the target column values. It contains columns that represent the conditions and the conditions' criteria in the decision tree, a prediction for each group, and prediction confidence. The Inbuilt Tree Diagram visualization can be used to visualize this decision tree.

The CARTree data set is outputted when you select these model and algorithm combinations.

Model	Algorithm
Numeric	CART for Numeric Prediction
Binary Classification	CART (Decision Tree)
Multi Classification	CART (Decision Tree)

Classification Report

This data set is a tabular representation of the accuracy metrics for each distinct value of the target column. For example, if the target column can have the two distinct values Yes and No, this data set shows accuracy metrics like F1, Precision, Recall, and Support (the number of rows in the training data set with this value) for every distinct value of the target column.

The Classification data set is outputted when you select these model and algorithm combinations.

Model	Algorithms
Binary Classification	Naive Bayes Neural Network Support Vector Machine
Multi Classification	Naive Bayes Neural Network Support Vector Machine

Model

Algorithms

Binary Classification

Naive Bayes

Neural Network

Support Vector Machine

Multi Classification

Naive Bayes

Neural Network

Support Vector Machine

Confusion Matrix

This data set, which is also called an error matrix, is a pivot table layout. Each row represents an instance of a predicted class, and each column represents an instance in an actual class. This table reports the number of false positives, false negatives, true positives, and true negatives, which are used to compute precision, recall, and F1 accuracy metrics.

The Confusion Matrix data set is outputted when you select these model and algorithm combinations.

Model	Algorithms
Binary Classification	Logistics Regression CART (Decision Tree) Naive Bayes Neural Network Random Forest Support Vector Machine
Multi Classification	CART (Decision Tree) Naive Bayes Neural Network Random Forest Support Vector Machine

Model

Algorithms

Binary Classification

Logistics Regression

CART (Decision Tree)

Naive Bayes

Neural Network

Random Forest

Support Vector Machine

Multi Classification

CART (Decision Tree)

Naive Bayes

Neural Network

Random Forest

Support Vector Machine

Drivers

This data set provides information about the columns that determine the target column values. Linear regressions are used to identify these columns. Each column is assigned coefficient and correlation values. The coefficient value describes the column's weight-age used to determine the target column's value. The correlation value indicates the relationship direction between the target column and dependent column. For example, if the target column's value increases or decreases based on the dependent column.

The Drivers data set is outputted when you select these model and algorithm combinations.

Model	Algorithms
Numeric	Linear Regression Elastic Net Linear Regression
Binary Classification	Logistics Regression Support Vector Machine
Multi Classification	Support Vector Machine

Model

Algorithms

Numeric

Linear Regression

Elastic Net Linear Regression

Binary Classification

Logistics Regression

Support Vector Machine

Multi Classification

Support Vector Machine

Hitmap

This data set contains information about the decision tree's leaf nodes. Each row in the table represents a leaf node and contains information describing what that leaf node represents, such as segment size, confidence, and expected number of rows. For example, expected number of correct predictions = Segment Size * Confidence.

The Hitmap data set is outputted when you select these model and algorithm combinations.

Model	Algorithm
Numeric	CART for Numeric Prediction

Residuals

This data set provides information on the quality of the residual predictions. A residual is the difference between the measured value and the predicted value of a regression model. This data set contains an aggregated sum value of absolute difference between the actual and predicted values for all columns in the data set.

The Residuals data set is outputted when you select these model and algorithm combinations.

Model	Algorithms
Numerics	Linear Regression Elastic Net Linear Regression CART for Numeric Prediction
Binary Classification	CART (Decision Tree)
Multi Classificatin	CART (Decision Tree)

Model

Algorithms

Numerics

Linear Regression

Elastic Net Linear Regression

CART for Numeric Prediction

Binary Classification

CART (Decision Tree)

Multi Classificatin

CART (Decision Tree)

Statistics

This data set's metrics depend upon the algorithm used to generate it. Note this list of metrics based on algorithm:

Linear Regression, CART for Numeric Prediction, Elastic Net Linear Regression - These algorithms contain R-Square, R-Square Adjusted, Mean Absolute Error(MAE), Mean Squared Error(MSE), Relative Absolute Error(RAE), Related Squared Error(RSE), Root Mean Squared Error(RMSE).
CART(Classification And Regression Trees), Naive Bayes Classification, Neural Network, Support Vector Machine(SVM), Random Forest, Logistic Regression - These algorithms contain Accuracy, Total F1.

This data set is outputted when you select these model and algorithm combinations.

Model	Algorithm
Numeric	Linear Regression Elastic Net Linear Regression CART for Numeric Prediction
Binary Classification	Logistics Regression CART (Decision Tree) Naive Bayes Neural Network Random Forest Support Vector Machine
Multi Classification	Naive Bayes Neural Network Random Forest Support Vector Machine

Model

Algorithm

Numeric

Linear Regression

Elastic Net Linear Regression

CART for Numeric Prediction

Binary Classification

Logistics Regression

CART (Decision Tree)

Naive Bayes

Neural Network

Random Forest

Support Vector Machine

Multi Classification

Naive Bayes

Neural Network

Random Forest

Support Vector Machine

Summary

This data set contains information such as Target name and Model name.

The Summary data set is outputted when you select these model and algorithm combinations.

Model	Algorithms
Binary Classification	Naive Bayes Neural Network Support Vector Machine
Multi Classification	Naive Bayes Neural Network Support Vector Machine

Model

Algorithms

Binary Classification

Naive Bayes

Neural Network

Support Vector Machine

Multi Classification

Naive Bayes

Neural Network

Support Vector Machine

Find a Predictive Model's Related Data Sets

Related data sets are generated when you train a predictive model.

Depending on the algorithm, related data sets contain details about the model like: prediction rules, accuracy metrics, confusion matrix, key drivers for prediction, and so on. These parameters help you understand the rules the model used to determine the predictions and classifications.

See What Are a Predictive Model's Related Data Sets?

On the Home page, click Navigator, and then click Machine Learning.
Click the Models tab.
Click the menu icon for a training model and select Inspect.
Click the Related tab to access the model's related data sets.
Double-click a related data set to view it or to use it in a project.

Add a Predictive Model to a Project

When you create a scenario in a project, you apply a predictive model to the project's data set to reveal the trends and patterns that the model was designed to find.

Note:

You can't apply an Oracle machine learning model to a project's data.

After you add the model to the project and map the model's inputs to the data set's columns, the Data Panel contains the model's objects, which you can drag and drop onto the canvas. Machine learning generates the model's values based on the visualization's corresponding data columns.

On the Home page, click Create, and then click Project.
Select the data set that you want to use to create the project and click Add to Project.
In the Data pane, click Add, and select Create Scenario.
In the Create Scenario - Select Model dialog, select a model and click OK.
You can only apply a predictive model. You can't apply an Oracle machine learning model.

If each model input can't be matched to a data element, then the Map Your Data to the Model dialog is displayed.
If the Map Your Data to the Model dialog is displayed, then in the Data Set field, select the data set to use with the model.
Match the model input and data elements as needed. Click Done.
The scenario is displayed as a data set in the Data Elements pane.
Drag and drop elements from the data set and the data model onto the Visualize canvas.
To adjust the scenario, right-click the scenario in the Data Elements pane and select Edit Scenario.
Change the data set and update the model input and data elements mapping as needed.
Click Save to save the project.