Conventional pattern classification aims to improve classification accuracy for the whole dataset. In the time of Big Data, however, there are circumstances in which people may take interest only in those typical instances and other issues like scalability and efficiency take priority. Keeping these issues in mind, in this paper, we revisit the perceptron algorithm. While it is a linear model, we show that with proper objective functions, it can be transformed to a probabilistic learner. The evaluation is carried out with the well known Pima diabetes database. The experimental results indicate that the perceptron algorithm is comparable to other sophisticated sophisticated algorithms in terms of the criteria discussed in this paper.
展开▼