Companies of all types and sizes are increasingly looking to predictive analytics to improve business decision making in Marketing, Risk Management, Fraud Detection, Network Security, Product Quality and many other areas.
Predictive modeling solutions now abound, but they all share three key weaknesses that makes it nearly impossible to align analytic models with business objectives.
Valuable events are rare, but the tools are biased to the majority.
The events the business wishes to predict for greatest return are often rare. For instance, respondents to a direct marketing campaign will be a tiny fraction of the targeted recipients. Such data is said to be highly unbalanced.
Conventional predictive modeling tools often have bias toward the majority class to minimize overall error rate, working against the business goal of predicting rare, rewarding events.
Event value varies widely, but the tools treat all events as the same.
The outcome of each individual business event contributes differently to the overall return from the correctly predicted events. For example, in a marketing campaign, one positive response will bring more revenue than another. Each event has a different cost and reward.
Conventional tools often treat each outcome equally, rather than considering its individual cost or reward, producing sub-optimal prediction result and therefore a sub-optimal business return.
Big Data modeling is compute intensive, but the tools don’t scale.
Big Data means more data types, more data points, more features, and more noise, which makes predictive modeling more complex and compute-intensive than ever.
Conventional tools are mostly built to run on desktop systems and individual servers. They cannot handle Big Data by taking advantage of the scalable compute power of clustered systems in a private or public cloud. This forces users to make tradeoffs between speed and accuracy which can further diminish the business return on prediction.
Unique Solution: Optimized Prediction
Yottamine Analytics offers a patent-pending predictive analytics solution that meets these challenges and produces models that deliver the highest possible business return.
Optimized Prediction combines revolutionary technology on three different levels to deliver a powerful, highly transparent modeling solution.
Intelligence of a new machine learning algorithm
The patent-pending Optimized Prediction algorithm is a machine learning breakthrough.
It’s built to handle unbalanced data, where target events, like marketing conversions and detectable frauds are rare and hard to predict.
It doesn’t just predict valid target events; it predicts the ones with the highest business value, such as the best up-sell targets or riskiest policies.
It doesn’t penalize dirty data or require extensive data tweaking. Big Data is real world data and Optimized Prediction takes it as it comes.
Power of highly parallel systems programming
Yottamine’s software uses highly parallel programming to ingeniously combines the CPU and memory resources of numerous networked servers to work as a single powerful system to build new predictive models.
The system is faster, easier to use, and more cost-effective than traditional desktop and server-based predictive modeling solutions. As a result, users can build more models and refresh them more frequently than they previously could.
In a recent performance benchmark, Yottamine produced a model with 95.75% accuracy in 90 minutes, while the most popular comparable product could only reach about 77% in more than six hours. See the details here.
Scalability of public or private cloud computing
Optimized Prediction runs in either the Amazon AWS public cloud or on-premises in a Eucalyptus private cloud. This gives the user smooth, automatic compute scaling to build models from the largest data sets.
Cloud scaling enables the modeler to work with Big Data without sampling and other reduction techniques. And, Yottamine’s cloud software is highly automated and doesn’t require any IT skills for configuration or operation.
In a recent Big Data prediction scale benchmark, Yottamine reached 97.82% accuracy in 360 minutes, while the competitor ran for 3 days before being terminated without producing any result. See the details here.
Provable Improvement in Business Return
Yottamine’s Optimized Prediction automatically enables better business return because it allows business value-driven learning objectives and doesn’t require artificially extended or enhanced data.
This was dramatically demonstrated in recent business case benchmark against a field of well-known solutions, from open source projects to commercial products from big software names.
Yottamine’s model produced nearly 10% more profit than the cup winner, but, more importantly, produced almost 30% higher profit per contact, and a 13-fold overall ROI for the campaign.
To put these results into perspective, The CMO Survey 2014 reports the average 2013 Marketing ROI for B2C product and services companies as 3.8%.
To learn how Optimized Prediction improves business ROI, download the white paper “Higher Business ROI with Optimized Prediction” here.
Enabling Technology: Yottamine Predictive Platform
Optimized Prediction runs on the Yottamine Predictive Platform, the result of five years of research and development in machine learning and software engineering dedicated to producing a fast, scalable, highly automated predictive modeling solution that is also easy to deploy and use.
Fastest, most scalable predictive modeling software available
YPP runs in the public Amazon AWS cloud or on-premises in a Eucalyptus private cloud and automates the processes of allocating and activating software and compute resources.
YPP combines highly parallel systems programming with advanced machine learning algorithms and flexible interfaces to create a highly scalable modeling server in the cloud.
The user, a data scientist or statistical data specialist uses either R scripts or Java programs to direct YPP to build models and integrate them into applications and data science workflows.
For an overview of the Yottamine Predictive Platform architecture, click here.
