Clustering is a machine learning method that can group similar data points. Mean Shift (MS) is a fixed window-based clustering algorithm, which calculates the number of clusters automatically but cannot guarantee the convergence of the algorithm. The main drawback of the Mean Shift Algorithm is that the algorithm requires to set a stopping criterion (threshold point) otherwise all clusters move towards one cluster and fixed bandwidth is used here. It cannot define the upper bound of iteration numbers and need to set the iteration numbers. This paper proposed a new Mean Shift Algorithm, called Improved Mean Shift (IMS) algorithm, which overcomes the all defined pitfalls of Mean Shift Algorithm. The IMS process KD-tree data structure was used to sort the dataset and all data points as initial cluster centroids without a random selection of initial centroids. In each iteration, it shifts the variable bandwidth sliding window to the actual data point nearest to the mean using k-nearest neighbours (kNN) algorithm and finds the number of clusters automatically. Also, this paper handles the missing values using Mean Imputation (MI). The IMS algorithm produces better results than the Mean Shift Algorithm on both synthetic and real datasets.

No abstract

No abstract

No abstract

No abstract