Predixion Software, founded in 2009 by Simon Arkell, Jamie MacLennan (ex-Microsoft) and Stuart Frost (ex-DATAllegro), is a relatively new arrival in the predictive analytics domain. It initially focused on providing self-service analytics in the cloud, although it now offers an on-premise version for organizations that need more direct control over analytic development.

In May, 2015, the company unveiled Predixion Insight 4.5, a new architecture (potentially quite innovative) whereby it embeds analytical algorithms directly into edge devices of an Internet of Things (IoT) system. By analytically-enabling edge devices, Predixion’s software allows organizations to detect problems in advance and send alerts. For example, this software could monitor the health of remote water pumping stations and issue alerts to a central control center if the software detects an impending failure or other state change like increased or unexpected vibration.

The key to Predixion’s IoT new architecture is a patent-pending software innovation called Machine Learning Semantic Model (MLSM). MLSM enables predictive analytics to easily be packaged and embedded into a variety of production environments including existing applications, databases, real-time streaming engines and even on connected or disconnected edge devices. What’s unique about Predixion Insight 4.5 is that the MLSM package is small enough to execute on a slow device with relatively little processing power.

With the addition of MLSM, Predixion Insight 4.5’s primary differentiators are:

1. Robust support for a variety of machine learning/predictive analytics algorithms.
2. Relatively easy to learn and use model development and deployment tools.
3. Analytics modules that can simply be deployed on IoT edge devices, on gateways and on analytics servers. By deploying close to IoT data sources, useful business insight is surfaced faster and at less cost.

Predixion Deployment Assistant view of an MLSM

IoT Technology like Predixion’s is broadly applicable across many industries including healthcare, manufacturing, energy and transportation. Predixion and other companies like it are well positioned to capitalize on the growing adoption of advanced analytics by maintenance repair and operations (MRO) service providers who maintain industrial equipment for companies in a variety of infrastructure-intensive industries: aerospace and defense, general transportation, resource extraction, energy production and distribution, and continuous and discrete manufacturing.

Most MRO providers have fixed-price support contracts, so their profit depends upon service efficiency. To improve the MRO bottom line, service providers must optimize deployment location and timeliness of spares inventory, service facilities and service personnel. Huge opportunities lie in the application of advanced analytics, using new big data sources and advanced predictive algorithms, to improve investments in working capital and maximize ROI.  Hence the value of Predixion and other similar advanced analytics providers including SAS, IBM, SAP, Microsoft, Dell, Fair Isaac, Oracle, KNIME, RapidMiner, etc.

Some of the interesting new features of Predixion Insight 4.5 are:

• The Deployment Assistant - A dialog-window-based tool that allows property-page-style configuration of modules for device, gateway or cloud. Users specify inputs, outputs and transformations, then export packaged code that can then be quickly embedded. The idea is to minimize time and errors encountered when packaging the algorithm code for run-time use.
• Additional algorithms - These new algorithms are size optimized for limited device memory. Roughly about 1MB is the smallest package using these new algorithms. These algorithms use ensemble methods and classification (decision forests with regression, boosted trees using a version of the Adaboost algorithm).
• Kura and ProSyst OSGi code module support - IoT run-times (e.g. device, gateway) that support OSGi can quickly/easily accept Predixion analytics modules.
• Runtime data shaping - Allows analytics developers to aggregate data (e.g. time series) to improve its value and meaning. This also reduces data flow rates to the model, improving overall information processing capacity.
• Integration - Wind River’s Intelligent Device Platform XT. This further simplifies gateway level predictive analytics deployment.