Predicting algal appearance on mortar surface with ensembles of adaptive neuro fuzzy models: a comparative study of ensemble strategies

• Published in: International journal of machine learning and cybernetics Vol. 10; no. 7; pp. 1687 - 1704
• Main Authors: Tran, Thu-Hien, Hoang, Nhat-Duc
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.07.2019; Springer Nature B.V
• Subjects: Adaptive systems; Algae; Artificial Intelligence; Artificial neural networks; Cement; Comparative studies; Complex Systems; Computational Intelligence; Construction; Control; Engineering; Facades; Fuzzy logic; Humidity; Literature reviews; Machine learning; Maintenance; Maintenance costs; Managers; Mechatronics; Mortars (material); Original Article; Pattern Recognition; Permeability; Precipitation; Research methodology; Robotics; Soft computing; Support vector machines; Systems Biology; ANFIS; Soft computing; Ensemble learning; Algal appearance; Machine learning; Building maintenance
• ISSN: 1868-8071; 1868-808X
• DOI: 10.1007/s13042-018-0846-1

Algal colonization on building façades causes paint defects, structural damages, and adverse effects on the health of occupants. Building managers need to be better informed about the status of the façade to develop effective maintenance programs. This study proposes a soft computing approach for predicting the appearance of algae on building façade with mortar cladding. The proposed approach is established by ensembles of Adaptive neuro fuzzy inference system, named as ANFIS-ESM, with random subspace (RSE), adaptive boosting (AdaBoost), random linear oracle (RLO), and stacking (STE) strategies. An experimental dataset which consists of 510 historical records and 12 conditioning factors has been collected for this research. Pre-training and post-training analyses on the importance of factors have also been performed by means of mutual information, ReliefF, and Fourier amplitude sensitivity test. Importantly, experimental results point out that STE strategy has attained the most accurate outcome with a classification accuracy rate (CAR) of 87.76% and an area under the curve (AUC) of 0.91. Moreover, the proposed ensemble framework has outperformed other benchmark models including the support vector machine, the random forest, and the rotation forest. Therefore, ANFIS-ESM is a promising tool to assist building managers in establishing maintenance programs.