A Global Clustering Approach Using Hybrid Optimization for Incomplete Data Based on Interval Reconstruction of Missing Value

Abstract

Incomplete data clustering is often encountered in practice. Here the treatment of missing attribute value and the optimization procedure of clustering are the important factors impacting the clustering performance. In this study, a missing attribute value becomes an information granule and is represented as a certain interval. To avoid intervals determined by different cluster information, we propose a congeneric nearest‐neighbor rule‐based architecture of the preclassification result, which can improve the effectiveness of estimation of missing attribute interval. Furthermore, a global fuzzy clustering approach using particle swarm optimization assisted by the Fuzzy C‐Means is proposed. A novel encoding scheme where particles are composed of the cluster prototypes and the missing attribute values is considered in the optimization procedure. The proposed approach improves the accuracy of clustering results, moreover, the missing attribute imputation can be implemented at the same time. The experimental results of several UCI data sets show the efficiency of the proposed approach.

Missing Value Imputation Designs and Methods of Nature-Inspired Metaheuristic Techniques: A Systematic Review

“…In FCM imputation with the PSO method [39]- [42], the missing values can be estimated from the observed data with different optimized weights to improve data quality. Recent work by Hu et al [43] presented missing values in hybrid numeric and granular forms.…”

“…Multi objective approaches Hybrid approaches GA [16]- [18] GP [19] PSO [20] MOGA-II [21]- [22] MOPSO [23] Bayesian ACO+ Bayesian [24] ABC+ Bayesian [25] Max-min ACO +bayesian [26]- [27] Bayesian+ tensor+chaotic PSO [28] Probabilistic GA+KNN [29] GMSA+MPSO+ WKNN [30] PSO+ covariance matrix [32] IDW+TR+ PSO [33] Clustering ACO+ clustering [34] FCM+GA [35][36] FCM+ SVR+GA [37] GA+SOM [38] FCM+PSO [39]- [42] GFM+PSO [43] PSO-ECM+ AAELM [44] ELM+PSO+ FCM [45] PSO+K-means+ ontology [46] SOM+FOA +LSSVM [47] DE+ clustering [48] GA+RF [49] GP+wrapper [55] [56] Neural network GSO+MLP [57] GA+MLP, SA+MLP, PSO+MLP, RF+MLP [58] SC-FITNET [59] SC-FDO+ MLP [60] DL-CS [61] DL-BAT [62] DL-GSA [63] PSO+LSVM [54] PSO+levy flight+SVM [53] MAIS+GA [50] GA+ARO [51] GP+tree vector [52] KNN+LAHC AWOA [31] The proposed approach enhanced imputation for missing multivariate data…”

“…Another standard metric is error for summarizing the performance of imputation and classification models. For example, researchers adopted error metric to measure the classification errors of the proposed imputation models in the missing iris dataset [22], [35], [39], poker hand dataset [24], [26], website phishing dataset [25] and health datasets [31]. On the other hand, RA is an indicator of how many estimations fail within a standard range [36], [37], [45], [61], [62], while specificity (also true negative rate) refers to the proportion of sample without the condition but obtained a negative result [18], [23], [28], [43].…”

Easy Particle Swarm Optimization for Nonlinear Constrained Optimization Problems

“…where ( ) j c x in (2) denote the problem-specific inequality constraints and ( ) j e x in (3) denote the problem-specific equation constraints. The problem-specific constraints (2)- (3) are derived from the limitations of resources in practical industries, such as in electromagnetics [2], real-time UAV path planning [3], prediction of seismic slope stability [4], energy development [5] [6], incomplete data clustering [7], and production inventory [8]. [9] presented a performance comparison of harmony search (HS), differential evolution (DE), and PSO for the standard benchmark functions.…”

Garra Rufa‐inspired optimization technique

“…The hybrid techniques can work in series, parallel, or with impeded calibration. [18][19][20][21][22][23][24][25][26][27] The particle distribution models can be classified into three major categories: one complex (OC-PSO), shuffled complex evolution (SCE-PSO), and random shuffled complex evolution (SCER-PSO). 28,29 The SCE-PSO modifications find a local optimum point instead of the global optimum point, while SCER-PSO divides the particles into sub-swarms according to a random replacement scheme.…”