Comparison of Machine Learning Functionalities of Business Intelligence and Analytics Tools

Abstract

Against the background of increasing need for data analysis in companies, which is carried out not only by data scientists, but also by end users with average IT knowledge, the question arises as to suitable BI systems that integrate machine learning functionality (ML) for data evaluation, but meet the requirements of an end user target group. This paper starts from the Gartner Magic Quadrant BI systems, which provide ML functionality in a free test license. For the five BI tools TIBCO Cloud Spotfire, SAP Analytics Cloud, Qlik Sense, Tableau and the Open Source tool RapidMiner, which is used as a neutral reference, the comparison is based on the ML workflow with the phases data acquisition and cleansing as well as model training, test, use and monitoring and analyzes the respective ML functionality offered. There are differences in the range of the spectrum of covered ML algorithm types classification, regression, clustering and association analysis, in the number of ML individual algorithms offered in each area and in the level of detail of the possible fine-tuning for the algorithms. The range extends from tools that are aimed at laymen without prior knowledge in data analysis and instead only contain a less transparent AI interface, over BI systems with ML algorithms that at least reveal the algorithm type, but hide the concrete algorithm and refrain from offering settings in order to address laymen with a basic understanding of data analysis, but little prior knowledge, to BI tools that boast a wide range of ML algorithms and a wealth of settings for fine-tuning and are therefore suitable for experts with sound specialist knowledge in the field of data science. In the end, fine-grained settings allow for more complex analyses, while BI tools with a high degree of automation and machine learning at the push of a button are more suitable for solving simple problems. It becomes clear that there is no such thing as a universally applicable BI system. Demands for detailed control options for the ML algorithms are diametrically opposed to the wide applicability by many users. Overall, the selection decision in the company must take into account the target user group and their prior knowledge in order to select the appropriate BI system.