Predictive analytics is a branch of advanced analytics that makes predictions about unknown future events. Historical data is collected to build a predictive model in order to predict the future. However, models can be outdated with time if it is not updated with new data that arrives. WSO2 SP addresses this issue by supporting online machine learning algorithms that enable you to evolve your models while they are alive. A key characteristic of an online predictive algorithm is that the memory and time requirements of the predictive analytics process do not grow over time.
Online predictive analytics for regression, classification, and clustering problems are supported via Siddhi queries. You can update the models by providing required data (and labels for the supervised case). You can also predict using the models by providing unseen data.
WSO2 SP keeps the state of the models in memory and persist the state time to time to recover from failures. The usage of system memory does not grow when new data points are detected. There are no great variations in the time taken to perform predictions and updates over time.
For detailed information about Machine Learning functionalities offered by WSO2 SP, see the documentation for the following extensions: