Application of ANFIS in the preparation of expert opinions and evaluation of building design variants in the context of processing large amounts of data

• Published in: Automation in construction Vol. 133; p. 104045
• Main Authors: Szafranko, Elżbieta, Srokosz, Piotr E., Jurczak, M., Śmieja, M.
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 01.01.2022; Elsevier BV
• Subjects: Adaptive neuro-fuzzy inference system; Adaptive systems; Algorithms; Artificial neural networks; Building design; Computer architecture; Construction investment variants; Criteria; CUDA multithread processing; Fuzzy logic; Multicriteria analysis; Construction investment variants; CUDA multithread processing; Adaptive neuro-fuzzy inference system; Multicriteria analysis
• ISSN: 0926-5805; 1872-7891
• DOI: 10.1016/j.autcon.2021.104045

There are many problems involved in the evaluation of variants of construction projects. One of the most difficult tasks is to establish evaluation criteria and assign them values, and afterwards to determine the degree to which the analyzed variants meet the criteria. This process is carried out based on opinions of experts, which usually constitute a large and heterogeneous set of data that is difficult to elaborate. As such, it raises most doubts and discussions among scientists. Therefore, the authors of this article propose to use a tool that can improve this process. The paper presents an example, and the Adaptive Neuro-Fuzzy Inference System (ANFIS) was used to solve the described problem. In the example, three variants were to be assessed and 17 criteria were used for the assessment, with the possibility of grouping them into four categories of main criteria. The developed ANFIS algorithm was implemented in the Compute Unified Device Architecture (CUDA) technology available in modern Nvidia graphics processors. The performance of the CUDA-ANFIS model was tested on several examples of real construction projects. It was found that the choice of the best investment option is not obvious when the final scores are the result of processing many thousands of data. In the evaluation process, fluctuations in the experts' responses may not only produce hard-to-distinguish final results, but may even reverse the order of the variants in the ranking. It has been shown that this will not happen if ANFIS is used to specifically filter both the input data and the intermediate results of the calculations. In addition, the use of CUDA technology speeds up calculations more than ten times. The new concept can be successfully applied to construct implicit decision models based on real data. [Display omitted] •A tool useful in determining values based on large databases, possibility to support expert assessments•Possibility to set weight values and assess the degree of compliance by variants•The system works on the raw data and eliminates subjective ratings•Efficient system that learns through repeated use•Increase in system efficiency with an increase in the number of processed cases