Gene-expression data obtained from the biological experiments always have thousands of dimensions, which can be very confusing and perplexing to biologists when viewed as a whole. Clustering analysis is an explorative data-mining technique for statistical data analysis that is widely used in gene-expression data analysis. Practical approaches employed for solving the clustering problem use iterative procedures such as K-means, which typically converge to one of many local minima. Here, we propose a simulated annealing approximation algorithm that is optimised using random walks to solve the K-means clustering problem. The algorithm is verified with synthetic and real-world data sets and compared with other well-known K-means variants. The new algorithm is less sensitive to initial cluster centres, and the primary strength of our algorithm is its ability to produce high-quality clustering results for thousands of high-dimensional data. However, the algorithm is computationally intensive.