Leveraging Classification Through AutoML and Microservices

Existing monolithic architecture used in ERPs (Enterprise Resource Planner) suffer from certain drawbacks such as increased complexity, increased startup time, reduced code readability and difficulty in debugging. This led to the emergence of micro‐service architecture which approaches software development as an amalgamation of loosely coupled, independently deployable micro applications (Micro‐Apps) of varied functionalities. These micro applications are privy to voluminous amount of data and incorporation of machine learning facility within them enables the users of the application to perform prediction of data circulated amidst the micro application. The aim of this work is to methodically obtain dataset and domain knowledge from consumers and make corresponding micro‐apps composed for them by micro intelligence application platform capable of classification or regression while simultaneously making sure that the complexities of ML code such as model selection, hyper parameter tuning are abstracted from the client side using AutoML. The user interface is developed using angular JS and it is integrated to the backend AutoML code using rest APIs. Further the machine learning model arrived is serialized and stored along with the micro app details in a database. The runtime interface of the micro‐application has a prediction service and is enabled to access the model that is deserialized and being run in a server. The functioning of proposed work is fully illustrated with the help of a three datasets namely Sonar, liver patients, and Insurance claim.