Fuzzy classification based on Combinative Algorithms

International Conference on Applied Sciences

自然科学（总论）

Saffie, N.A.M.^1 ; Rasmani, K.A.^1 ; Sulaiman, N.H.^2

Faculty of Computer and Mathematical Sciences, Universiti Teknologi MARA Seremban Campus, Negeri Sembilan, Seremban

70300, Malaysia^1

Faculty of Computer and Mathematical Sciences Universiti Teknologi MARA, Selangor, Shah Alam

40450, Malaysia^2

关键词: Classification accuracy;    Classification algorithm;    Classification models;    Classification tasks;    Classification technique;    Fuzzy classification;    Multiple-model approaches;    Ordered weighted averaging operator;   

Others  :  https://iopscience.iop.org/article/10.1088/1757-899X/477/1/012026/pdf
DOI  :  10.1088/1757-899X/477/1/012026

学科分类：自然科学（综合）

来源: IOP

【 摘 要 】

Classification models are used in order to predict the class label of objects for which the class label is unknown. The performance of a single classification model can be determined on the basis of the classification accuracy. However, it is difficult to determine which single-model is the best classification technique in a specific application domain since a single learning algorithm may not uniformly outperform other algorithms over various datasets. Furthermore, most classification algorithms, using either fuzzy or non-fuzzy approaches, produce results in the form of crisp or categorical classification outcomes. Unfortunately, in certain applications, the classification outcomes that represent the class labels may involve categorisations which are fuzzy in nature. In this study, a new technique is developed to handle the classification task based on combinative algorithms by adopting multiple-model concepts. The aggregation process is based on a modified fuzzy Ordered Weighted Averaging (OWA) operator. The performance of the proposed technique has been analysed using the Domestic Water Consumption (DWC) dataset obtained from a previous study. The findings show that the proposed method consistently outperforms all selected single-models and performs better than a statistical multiple-model approach, known as the Pool Method. This result indicates that the proposed technique has the potential to assist decision-making, especially in the following situations: i) when it is difficult to decide which model is better for a specific classification problem and, ii) to decide selected unconfirmed cases.