About Logi Predict

The Logi Predict add-on module for Logi Info allows users to analyze historical or transactional data and make statistical predictions about current data.

About Predictive Analytics

Predictive Analytics is a statistical technique used to make predictions about future events. Typically, it discovers patterns found in historical and transactional data and applies them to current data in order to produce a probability score. This score allows us to predict outcomes.

One well-known application of this is consumer credit scoring, which is widely used throughout the financial services industry. Scoring models process a customer's credit history, personal data, loan application, etc., in order to predict the likelihood that the customer will make future payments on time.

Predictive Analytics is distinguished from simple numerical forecasting by the level of data granularity it brings to the predictive process. It is said to learn from the historical data, rather than just crunch it through calculations.

The opportunities to usefully employ Predictive Analytics in business are almost limitless. Prediction examples include the credit scoring discussed earlier, hospital re-admissions, software bugs, customer "churn", and many more applications of the technology.

Back to top

About Logi Predict

Logi Predict is an add-on module that enables special elements in Logi Info and adds a pre-built Logi application to your computer.

It leverages the power of the Logi Info platform and adds predictive technology,

The included Logi Info application, also called Logi Predict, lets regular users make accurate predictions, without the need for data scientists. It can be used "as-is", right out of the box, and it can also be customized and branded to fit your needs.

Logi Info makes it easy to embed this capability into your other applications, and even includes a REST API so that other applications can get prediction results. It's also easy for end-users to interact with the results, and easy to "operationalize" the predictive analytics execution and maintenance.

Logi Predict uses the "R" statistical analysis environment to process data and offers several different analytic methods for prediction processing.  

The Predictive Process

The two-step process used by Logi Predict is shown in the following diagram: 

In Step 1, historical or transactional data is processed using predictive algorithms to create a "prediction model", a process referred to as training a model. The model contains information about the patterns and other statistical indicators discovered during its analysis of the data. Logi Predict offers four types of models: Classification, Clustering, Forecast, and Outliers, which are discussed later on. A model only includes the results of its analysis of the historical data, not the historical data itself. Once created and trained, models are stored and can be re-used.

In Step 2, the model is applied to the "new" data, the data about which we want predictions, by creating a "prediction plan". The model applies the patterns and indicators it learned in Step 1 to the new data, resulting in predictions. This step can be scheduled to run repeatedly against new or changing data over time.

Back to top

Prediction Model Types

Logi Predict offers these prediction model types: 

Model Type Use This Model To Description


Predict categorical values

This model lets you categorize new data points. Use this model for questions with binary (Yes/No) answers, like these:

  • How likely is this customer to churn?
  • Will a user click on this link?
  • Will a user buy this product?
  • Will this loan be approved?
  • Is this a fraudulent transaction?
  • Is this employee at risk of leaving?
  • Is this a low or high risk claim?

Predict numeric values

This model lets you predict the possible value of a metric. Use this model for questions with numeric answers, like these:

  • What will Call Center case volume be tomorrow? Next week, next month?
  • How long will it take to perform this job?
  • How many customers would come to my shop?
  • What will the response time be for this issue?
  • What would the emergency wait time be?
  • How many watches should I keep in inventory?
  • How many customers will default?

    Time Series

Predict time series data

This model lets you predict future numeric values using historical time series data. Use this model for time-related questions, like these:

  • Forecast units sold for the next few weeks.
  • Forecast the birth rate at hospitals for the next few months.
  • What will be the closing price of a stock for the next few days?
  • How many passengers will pass through a train station tomorrow?
  • What will be the unemployment rate for a state in the next few quarters?


Find smart data groupings

This model lets you gather data points into smart groups or segments based on their attributes. Use this model to:

  • Group customers into smart buckets based on buying patterns and demographics.
  • Group loans into smart buckets based on loan attributes.
  • Group SaaS customer data into groups to understand global patterns.
  • Identify groups of insurance policy holders.



Find outlier values

This model lets you identify values that are anomalous, or outside the expected range of values. Use this model for questions with binary (Yes/No) answers, like these:

  • Is this a fraudulent claim?
  • Do sensor measurements indicate an anomaly?
  • Are call volumes out of the ordinary?
  • Has application response time changed?

Selecting the correct model type is important and is driven by the questions you want answered or the kinds of results desired. The choice of model type determines which predictive algorithms will be used to train the model. 

A Library of Models

Logi Predict lets you create a "library" of models for use in prediction plans. This allows models to be used, re-used, and even updated, saving you time and effort. This also allows individual models to be used in multiple prediction plans. Once defined, models are stored by Logi Predict for evaluation, assignment to prediction plans, and management.  

Selecting a Prediction Algorithm

Logi Predict offers these prediction algorithm choices: 

Model Available Algorithms Characteristics


  • Generalized Linear Model for Two Values
  • Gradient Boosted Model
  • Random Forest

 Faster training, less accuracy
 Fast ensemble training, high accuracy, class accretion
Fast ensemble training, high accuracy, decision trees


  • Gradient Boosted Model
  • Linear Regression
  • Random Forest
Fast ensemble training, high accuracy
Faster training, less accuracy
Fast ensemble training, high accuracy
Time Series
  • Prophet

High-quality forecasts for time series data that has multiple seasonality with linear or non-linear growth


  • K-Means

 High accuracy


  • All Together
  • Column by Column

Selection is based on number of variables

When creating a model, how do you know which algorithm to select? The answer to the question depends on considerations such as:

Accuracy - Accuracy may not be a priority and sometimes an approximation is good enough. If so, you may be able to reduce training time significantly using more approximate methods.

Training Time - The amount of time needed to train a model varies depending on the algorithm used. Some algorithms are more sensitive to the number of data points available than others and work better with large data sets.

Linearity - Linear classification algorithms assume that classes can be separated by a straight line, while linear regression algorithms assume that data trends follow a straight line. These assumptions are useful for some predictions, but bring accuracy down for others. Nonetheless, linear algorithms are often a first choice because they lack complexity and provide fast training.

Ensemble Training - Algorithms can provide "ensemble" training, which means that they generate hundreds of decision trees and averages their results during the training process. This "smooths" the results, so outlier data points have less effect, and improves accuracy.

Number of Parameters - Algorithm parameters, such as number of decision trees, number of clusters, number of rows of data to use for training, affect an algorithm's behavior. Training time and accuracy can be quite sensitive to parameter settings. Typically, algorithms with a large number of parameters require the most experimentation in order to find the best parameter values.

 Generally, if you have the time and no other considerations prevent it, try to train and test models with all the algorithms available.

Back to top

Prediction Plans

In order to produce predictions, you execute a "prediction plan". This plan specifies all of the settings needed to create the results, including:

  • The prediction model to be used
  • The data to be processed using the model
  • The details of the database table where the results will be stored, or other output options
  • The schedule for executing the plan (immediate or scheduled)

Once defined, prediction plans are stored by Logi Predict for execution, evaluation, and management.

Back to top

Example Use Cases

Here are some example use cases for this technology, drawn from real-life data: 

Predict Hospital Readmissions

Readmission of previously discharged patients is viewed as one measure of healthcare success and a high readmission rate can have financial penalties. Hospitals want to identify, through prediction, those patients who have a high propensity to be readmitted within 30 days.

Using actual patient data and a Classification model, Logi Predict had an 89% accuracy rate with this prediction. This can allow hospitals to assess, before discharging a patient, what their readmission risk would be and to take appropriate action. 

Predict Monthly Payment Default

Banks want to minimize risk by proactively identifying customers who could default next month. This helps the bank to take preventive action by potentially engaging with the borrower to help them make the best financial decisions. It also helps the bank manage their financial risk exposure.

Using bank data and a Classification model, Logi Predict had an 82% accuracy rate with this prediction. This was a complex prediction plan that used Logi Predict's "ensemble learning" capability to combine the results of several models. Based on the prediction, banks can create a campaign to engage with customers with a high default risk to negotiate a new payment plan. 

Predict Customer Churn

New customers are expensive to acquire so companies focus on making existing customers happy and keeping them engaged. Most telecom companies, for example, have a customer attrition, or "churn", rate ranging from 3% to 9% which results in $100Ms of lost revenue. These companies want to identify customers they may lose, so they can take preventive action.

Using telecom company data and a Classification model, Logi Predict had an 80.5% accuracy rate with this prediction. The results could be used to run retention campaigns that help reduce customer churn. 

Here are some additional use cases commonly identified by users:

  • Marketing - Predict propensity to click a web page link
  • Retail - Predict demand for specific products during the Christmas holiday season
  • Insurance - Identify fraudulent claims
  • Human Resources - Predict likelihood of an employee leaving the company
  • Call Center - Predict call volume for the next month
  • Software Testing - Predict length of time testing will require 

There are, as we've seen, many interesting questions that can be answered by Logi Predict.

Back to top

What to Read Next

Now that you understand the basic information about using Logi Predict, you can continue to reading with these related topics:

For End-Users For Technical Staff & Administrators For Developers

Back to top

Release Pairings

For more information about which Logi Info and Discovery Module versions should be used together, see Introducing Add-on Modules. Also, refer to the Release Notes for each Logi Info version.

Back to top