NEC Corporation has announced the development of a “Predictive Analytics Automation Technology” that completely automates the process for large-scale data predictive analytics performed by relational databases that are widely used for business systems.
Currently, when analyzing relational databases composed of multiple databases, a great deal of work is required for processes that include the discovery and association of complex relationships between databases by skilled scientists, as well as the adjustment of prediction models by machine learning. Moreover, there is a shortage of well-trained data scientists capable of handling the rapidly growing need for advanced data analysis. As a result, there is heavy demand for highly accurate analysis methods that quickly perform large-scale data analysis and are user friendly for non-experts.
NEC’s Predictive Analytics Automation Technology, developed as part of its cutting-edge portfolio of artificial intelligence (AI) technologies, NEC the WISE, automates the series of processes for predictive analysis, from the extraction and design of a data item (feature that is effective for analysis, to the creation of the most suitable predictive model. As a result, even if an individual lacks advanced data analysis skill, it is possible to perform predictive analysis in a short time that is equal to or better than the accuracy of a well-trained data scientist.
Joint trials carried out by Sumitomo Mitsui Banking Corporation and NEC confirmed that this technology successfully maintained accuracy and reduced predictive analysis to just one day, in comparison to conventional methods that require 2-3 months of work by a professional analyst.
“This new technology can contribute to the acceleration of business decisions, including strategic planning, hypothesis verification and policy implementation, based on the discovery of new potential needs,” said Akio Yamada, general manager, Data Science Research Laboratories, NEC Corporation. “We aim to provide this technology as a service within the 2017 fiscal year for companies seeking to independently perform effective big data analysis.”
Among the features is the “Automatic Feature Design Technology” which automatically discovers feature quantities for relational data bases. This technology strengthens the “Automatic Feature Design Technology” that NEC announced in 2015 and automatically designs the feature for the relational databases that are widely used for business systems.
Based on the relationship of multiple databases, AI searches for and discovers hypotheses at high speed for combinations of data items (feature quantities) that are effective for prediction. Moreover, the system automatically creates the large number of queries to generate features from the databases.
As a result, the time and labor for analysis is significantly shortened since neither large amounts of work are necessary for feature hypothesis planning, which is dependent on analysis experience and knowledge about data, nor are database operations required for creating feature quantities. Furthermore, in comparison to manual analysis, a great deal more hypothesis searching can be executed in a short time, more accurate analysis results can be achieved, and new findings that may not have been noticed by manual processes may be discovered.
Another feature is the “Automatic Prediction Model Design Technology” which enables the automatic design of the most suitable model for the data. Based on feature data, a wide range of prediction models are created using various machine learning methods, such as NEC’s “Heterogeneous Mixed Learning,” logistic regression and decision trees. The prediction model that provides the most suitable analysis results for a user’s goals is selected. Reasons for the predicted value calculated by the prediction model are also provided.
Since users are able to understand the basis of the prediction, they are also able to make the most suitable judgement and implement the most appropriate plan in response to a situation.
NEC also developed a Graphic User Interface (GUI) for intuitive operation, where a display provides users with easy to understand instructions to search for feature quantities and create predictive models.