Co-active neuro-fuzzy inference system model as single imputation approach for non-monotone pattern of missing data

Data imputation aims to solve missing values problem which is common in nowadays applications. Many techniques have been proposed to solve this problem from statistical methods such as Mean/Mode to machine learning models. In this paper, an approach based on Co-active Neuro-Fuzzy Inference System named CANFIS-ART is proposed to automate data imputation procedure. This model is constructed from the Neural Network adaptative capabilities and fuzzy logic qualitative approach using the Fuzzy-ART algorithm. Performance of CANFIS-ART model is compared to other state-of-the-art imputation techniques such as Multilayer Perceptron or Hot-Deck, among others, using a total of eighteen databases exposed to a perturbation procedure based on the random generation of non-monotone missing values pattern. The data sets cover a wide range of fields, types of variables and sizes. A comparison of databases imputed by these models using a set of three classifiers has been conducted. A statistical analysis of these results employing Wilcoxon signed-ranked test has been included. Experiments show that CANFIS-ART approach not only outperforms these state-of-the-art techniques but also demonstrates a higher level of generalization capability, increasing the data quality contained in databases with missing values.