New approaches to normalization techniques to enhance k-means clustering algorithm

• Main Authors: Dalatu, Paul Inuwa, Midi, Habshah
• Format: Article
• Language: English
• Published: Institute for Mathematical Research, Universiti Putra Malaysia 2020
• Online Access: http://psasir.upm.edu.my/id/eprint/38339/1/3.%20Paul%20n%20Habshah.pdf; http://psasir.upm.edu.my/id/eprint/38339/; http://einspem.upm.edu.my/journal/fullpaper/vol14no1jan/3.%20Paul%20n%20Habshah.pdf

Clustering is fundamentally one of the leading origin of basic data mining tools, which makes researchers believe the normal grouping of attributes in datasets. The main aim of clustering is to ascertain similarities and arrangements with a large dataset by partitioning data into clusters. It is important to note that distance measures like Euclidean distance, should not be used without normalization of datasets. The limitation of using both Min-Max (MM) and Decimal Scaling (DS) normalization methods are that the minimum and maximum values may be out-of-samples when dataset are unknown. Therefore, we proposed two new normalization approaches to overcome attributes with initially large magnitudes from overweighing attributes with initially smaller magnitudes. The two new normalization approaches are called New Approach to Min-Max (NAMM) and New Approach to Decimal Scaling (NADS). To evaluate the performance of our proposed approaches, simulation study and real data applications are considered. However, the two proposed approaches have shown good performance compared to the existing methods, by achieving nearly maximum points in the average external validity measures, recorded lower computing time and clustering the object points to almost all their cluster centers. Consequently, from the results obtained, it can be noted that the NAMM and NADS approaches yielded better performance in the data preprocessing methods, which down weight the magnitudes of large values.