The use of a consumer-grade Brain-Computer Interface (BCI) has seen significant interests among researchers and hobbyists like communities. It has been suggested as a viable mean to control robots, improve learning experience and even to classify thought patterns. This paper investigates the possibility of using the NeuroSky Mindwave headset, a very cheap and popular single electrode BCI, for such endeavors by means of unsupervised machine learning algorithms. Firstly, the raw Electroencephalography (EEG) signals from 10 different subjects were acquired while they performed various mental activities. The mental activities ranged from listening to relaxing music to doing mathematical calculations. Secondly, the EEG signals were filtered to obtain the Gamma, Beta, Alpha, Theta and Delta brainwaves. Finally, k-means, fuzzy c-means and Self-Organizing Maps (SOMs) clustering algorithms have been applied to group the brainwaves according to their similarities. The performance of the cluster algorithms was benchmarked using distance metric maps, cluster silhouettes, Calinski-Harabasz index and Davies-Bouldin index. K-means clustering algorithm has showed some power of separating different mental activities into groups. The minimum Mean Silhouette Value has been found to be 0.475 when the number of clusters is 3 and the highest CH-index registered has been 65.7. These results show an interesting possibility for using the MindWave headset in applications where the number of mental activities to be harvested may not be greater than 2 or 3 at most.