Enhancing K means by unsupervised learning using PSO algorithm

Data Clustering in Data Mining is a domain which never gets out of focus. Clustering a data was always an easy task but achieving the required accuracy, precision and performance was never so easy. K means being an archaic clustering algorithm got tested and experimented thousands of times with variety of datasets and other combination of algorithm due to its robustness and simplicity but what this algorithm proposed was not suggested before. It used K means algorithm for the evaluation and validation purposes whereas optimization of the data is done with the help of Particle Swarm Optimization Algorithm. The drawbacks of K means mainly its local convergence property and initializing number of clusters at an early stage has aroused the process of working on this algorithm. So, for attaining the global convergence the Swarm Intelligence is preferred over Genetic Algorithm and many other techniques and for the latter one we combined two functions one of them helps in knowing the number of clusters which are optimal for the particular dataset and the other one validates the results using another function and compares the various metrics which will define the goodness and fitness of an algorithm. In one line the complete overview of the proposed algorithm can be described as ‘Evaluating the data using an Evalcluster Function, performing Validation with the help of an Evaluate Function of the K means and giving the final touch of Optimizing the data by K means PSO Algorithm’. The algorithm is tested for over 4 datasets available in UCI Repository and the results were unexpectedly great.