Create, deploy, and maintain analytic applications that engage users and drive revenue. See a Logi demo

Logi News

Logi Predict Product Update: Time Series Forecasting, Data Transformations, and Imbalanced Data Optimization

By Sriram Parthasarathy | May 13, 2019
Share on LinkedIn Tweet about this on Twitter Share on Facebook

Our development team has been hard at work on some exciting new features for Logi Predict—the only predictive analytics solution designed specifically to embed inside existing applications. Logi Predict makes it easy for application teams, product managers, and developers to embed advanced analytics capabilities that highlight future trends to improve operations within the context of the application.

The latest release of Logi Predict 3.0 includes new a time series prediction model, algorithm-assisted data wrangling features, new ways to handle imbalanced data (one sided), and the ability to write predictions to a different database.

>> Related: Meet Logi Predict <<

Time Series Forecasting with Daily, Weekly & Monthly Seasonality

A time series is a sequence of data points that are captured in a timely order. For example, the number of daily calls received in the past three months; sales for the past 20 quarters; and the number of patients who showed up in a given hospital in the past six weeks. Time series forecasting is an important area of machine learning, since many predictive analytics problems involve a time component.

Time series forecasting was one of the most requested features for Logi Predict. Now you can start building time series prediction models without writing any code and embed them in your reports and dashboards. We used the Prophet algorithm to power robust time series data in Logi Predict 3.0. Prophet is based on an additive model where non-linear trends are fit with yearly and weekly seasonality, plus holidays. Unlike other time series forecasting algorithms, Prophet is good at handling missing data, large outliers, and large datasets quickly.

Why use time series forecasting? Not all growth is linear or static. You may want to better model exponential growth and better align the model to your trend. Or you may need to forecast for multiple projects or multiple regions at a time, instead of just one at a time. The time series forecasting in Logi Predict helps you do it all.

What’s a real-world example of time series forecasting for predictive analytics? Say you are trying to predict sales for the next three months, or forecast how many customers will show up in your bank the next few days. That is where the new time series model helps you forecast the future.

 

 

Smart Data Transformations

We have yet to see a dataset that’s perfectly clean. And that’s okay. Logi Predict 3.0 has a number of features to help you transform data.

First, you can now get a good picture of what is in your data column. You can also control how to handle outliers, missing values, and (most popular with our users) creating statistical bins or quantiles.

The latest release also gives you granular control and visual column stats with actions. Now, anyone using Logi Predict can see column stats and take action based on what the data says.

 

Easily Handle Many Unique Values

If you’ve ever created a predictive model with lots of levels, you know how difficult it is to handle. Many machine learning algorithms can’t handle categorical variables. Even for those that can, categorical data can pose a serious problem if they have high cardinality, meaning too many unique values. It is not only hard to use high-cardinality columns while building models, but it is also very taxing on computer resources. 

We’ve made it easier to handle high-cardinality columns such as states, suppliers, zip codes, and partners. When you build predictive models, Logi Predict automatically takes care of high cardinality columns. No need to transform them—just pass them on as it is. (I can hear a sign of relief from the developers!)

 

Managing Imbalanced Outcomes

Not all outcomes are perfectly balanced. For example, consider a hospital that’s trying to predict if a patient needs to be screened for cancer. It’s a highly imbalanced problem, since maybe only 5 percent of patients will need to be screened. If not done correctly, your algorithm might say no one needs to get screened—and it will be right 95 percent of the time. But the 5 percent it misses is awful.

Logi Predict now comes with class imbalance optimization to help you handle these cases.

 

Keep Test Data and Production Data Separate

Understandably, predictive analytics users do not want to write-back the predictions to the same database where their source data is. You want to read from the live application data from one database and do the predictions and write-back the predicted results to a different database. Now, Logi Predict ensures you can write-back predictions to a new database that’s different from the source data we’re using to predict.

 

Want to experience Logi Predict 3.0 yourself? Watch the free demo today >

About the Author

Sriram Parthasarathy is the Senior Director of Predictive Analytics at Logi Analytics. Prior to working at Logi, Sriram was a practicing data scientist, implementing and advising companies in healthcare and financial services for their use of Predictive Analytics. Prior to that, Sriram was with MicroStrategy for over a decade, where he led and launched several product modules/offerings to the market.

Subscribe to the latest articles, videos, and webinars from Logi.