Min max kurtosis distance based improved initial centroid selection approach of K-means clustering for big data mining on gene expression data

Genome clustering is one of the big data applications that identify the prognosis of terrifying diseases and biological processes across enormous sets of genes. The K-Means (KM) algorithm is the most commonly used clustering algorithm for gene expression data that extracts hidden knowledge, patterns and trends from gene expression profiles for decision-making strategies. Unfortunately, the KM algorithm is extremely sensitive to initial centroid selection since the initial centroid of clusters influences computational effectiveness, efficiency, cost and local optima issues. The existing initial centroid initialization algorithm attains high computational complexity due to extensive iterations, distance computation, data and result comparison on high dimensional data. To overcome these weaknesses, this study suggested the Min–Max Kurtosis Distance (MMKD) algorithm for big data clustering in a single machine environment. The MMKD algorithm resolves the KM clustering weaknesses by the distance between data points of origin and minimum–maximum kurtosis dimension. The performance of the proposed algorithm is compared to KM, KM++ , ADV, MKM, Mean-KM, NFD, K-MAM, NRKM2, FMNN and MuKM algorithms by internal and external effectiveness validation metrics with efficiency measurement on sixteen gene expression datasets. The experimental evaluation demonstrates that the MMKDKM algorithm reduces iterations, local optima, computation costs, and improves cluster performance, effectiveness and efficiency with stable convergence than other algorithms. The statistical analysis of this study promised that the proposed MMKDKM algorithm achieves a significant difference.