MILKDE: A new approach for multiple instance learning based on positive instance selection and kernel density estimation

Multiple Instance Learning (MIL) is a recent paradigm of learning, which is based on the assignment of a single label to a set of instances called bag. A bag is positive if it contains at least one positive instance, and negative otherwise. This work proposes a new algorithm based on likelihood computation by means of Kernel Density Estimation (KDE) called MILKDE. Using the LogitBoost classifier, its performance was compared to that of forty-three MIL algorithms available in the literature using five data sets. Our proposal outperformed all of them for the Elephant (87.40%), Fox (66.80%) and COREL 2000 data sets (77.8%), and achieved competitive results for the MUSK 1 (89.20%) and MUSK 2 (87.50%) data sets, which are comparable to the higher accuracies obtained by other methods for this data sets. Overall results are statistically comparable to those obtained by the most well known methods for MIL described in the literature.