
Machine Learning
Machine Learning
Machine learning applies statistical models to the insurance data you have in order to make smart predictions about the future data. Those predictions can help you find patterns and signals in the noise and extract value from all the data you’re collecting. Machine learning can mine vast swaths of data in minutes, find and analyze the patterns, and make predictions in ways that no human analyst could. From the various machine learning methods employed to analyze insurance data, including support vector machines and neural networks, we have chosen decision trees due to the higher accuracy of results and the ease of presentation.
Classification and Regression Trees (CRT), a recursive partitioning method, builds classification and regression trees for predicting continuous dependent variables (regression) and categorical predictor variables (classification). Other tree building algorithms are QUEST (Quick, Unbiased, Efficient Statistical Trees) and CHAID (Chi-square Automatic Interaction Detector).
kSpheres
The kSpheres wizards enable the actuaries to configure the data sources to be mined for the model. In the initial stage of deployment, the configuration of data sources and management of ontologies will be jointly accomplished in the PREDMINE back office. The future intention however is that the users themselves be empowered to do so in a seamless Web-enabled UX if they chose to do so. Hence the architecture must be capable of seamless UX is crucial in the future.
The kSpheres platform periodically mines the configured sources for all PREDMINE customers to deliver accurate and enriched data to the PREDMINE server for predictive analysis. The core architecture of the kSpheres platform has already been developed but its functionality must be extended to support a broader spectrum of internal and external sources, more advanced multi-lingual text analysis and a greater big data capacity. kSpheres Plugins much of the text analysis is accomplished by ‘best-of-breed’ plug-ins to cater to a broad spectrum of languages and special needs. A core set of plugins must be chosen with provisions for more in a manner that will automatically train and execute appropriate plugins for each data sources. The number and nature of sources may grow with each customer so the plugin subsystem must be scalable.
kSpheres machine learning enables the app to automaticallylearn new entities (e.g. cars and properties) as they surface in the data mining. Future needs might require that data be pushed rather than pulled. The semantic nature of kSpheres enables the app to interact with the mined data in manners beyond a conventional database, e.g. semantic parametric search and SPARQL for inference.