In the digital era, recognition of handwritten Kannada characters and words and converting them to machine editable form is a challenging research. It has applications like reading aid for blind, digitization of the handwritten documents, postal mail segregation and author, and handedness identification in forensics. The aim of this paper is to build the dataset of handwritten Kannada vowels using unsupervised learning technique. To build the dataset by moving the characters to their respective folders which is done by feature extraction methods like Histogram of Oriented Gradients (HOG), Run Length Count (RLC), Chain Code (CC), Local Binary Pattern (LBP). Classification of characters is achieved with clustering, an unsupervised learning method. To achieve a higher recognition rate, combination of feature extraction methods have been applied. The combined features of HOG and RLC yielded the best recognition rate of 86.92%. Experiment is carried out with the data collected from 100 people.