Clustering applications in data sets with missing data

Missing data is a possible case in studies. Clustering analysis is a method that using to find natural groups of objects. Cluster analysis in many areas is one of the most used approaches in data analysis. In the analyzed datasets, there may be missing data at various rates. There is no definite information about which is the best in the clustering methods that can be used for the analysis of data sets containing lost data, which is completely random in numerical data. According to the number of data and data structure, each method has its advantages and deficiencies. In this study, clustering of data with continuous full and lost data is examined. Clustering analyzes were performed using k-ortalama clustering from division clustering techniques, self organization map (SOM) and learning vector quantization (LVQ) methods from artificial neural network-based clustering techniques, and the results obtained were compared. In order to make a qualitative comparison, eleven sets of real data, which are widely used in making such comparisons, were used in the literature. According to the results of the analysis, it is seen that as the loss data rate increases in all methods, the correct classification rates decrease and the SOM clustering method performs better than the k-averages and LVQ methods in seven data sets. In four data sets, it was determined that LVQ performed better depending on the structure of the data. In this study, the best method has been proposed by presenting an alternative to working with missing data within the stated limitations.