An Architecture for Simplified and Automated Machine Learning

Abstrak

learning has been adopted by businesses to analyze their vast data in order to make strategic decision. However, knowledge in machine learning and technical skill are usually required to prepare data and perform machine learning tasks. This obstacle prevents smaller businesses with no technical knowledge to utilize machine learning. In this paper, we propose an architecture for simplified and automated machine learning process currently supporting the data classification task. The architecture includes a method for characterizing datasets, which allows for simplifying and automating machine learning model and hyperparameter selection based on historical execution configurations. Users can simply upload their datasets via a web browser, and the system will determine the possible models and their hyperparameter configurations for the users to choose from. The prototype shows the feasibility of the proposed architecture. Although the accuracy is still limited by the small execution history and the cleanliness of the input datasets, the architecture can minimize user involvement in the machine learning process so that non-technical users can perform data classification through a web browser without installing any software.