Uncertainty Based Optimal Sample Selection for Big Data

In Machine learning and pattern recognition, building a better predictive model is one of the key problems in presence big or massive data; especially, if that data contains noisy unrepresentative samples. These types samples adversely affect may degrade its performance. To alleviate this problem, sometimes, it becomes necessary to sample after eliminating unnecessary instances by maintaining underlying distribution intact. This process called sampling instance selection (IS). However, process, substantial computational cost involved. paper discusses an uncertainty based optimal (UBOSS) method which can select subset efficiently. Our proposed work comprises three main steps; initially, uses IS identify patterns representative from original set; then, uncertainty-based selector designed obtain fuzziness (i.e., type uncertainty) those using classifier whose output membership fuzzy vector; further utilizes divide-and-conquer strategy Experiments are conducted on six datasets evaluate performance method. Results show our methodology outperforms when compared with optimum samples) baseline methods CNN, IB3, DROP3).