Novel approach for incremental learning using ensemble of SVMs with particle swarm optimization

Problems of online learning, or real-world scenarios where machines are required to be trained on batches of data instead of the entire superset, necessitate an efficient and accurate incremental learning approach which both, consumes less time and is computationally light; Incremental Support Vector Machines (SVM) has proved to be a promising panacea. We propose a novel approach, which firstly uses SVMs in an ensemble manner for learning from the batches of data, it then discards the correctly classified data and trains a new SVM on the misclassified data points, the weighted sums of the correctly trained SVMs, and the machines trained on the misclassified points, are then used to obtain the final classification values. Weight assignment to SVMs, for final classification at each step, is done using Particle Swam Optimization (PSO). The simulation results attest to the veracity of the model, and comparisons suggest that the proposed approach holds promise.