MBFerns: classification and extraction of actionable knowledge using Multi-Branch Ferns-based Naive Bayesian classifier

Classification is one of the tasks that are most frequently carried out in real world applications. A large number of techniques have been developed based on statistics and machine learning methods. These classification techniques usually suffer from various limitations, and there is no single technique that works best for all classification problems. Two major drawbacks in existing techniques are accuracy and lack of actionable knowledge from results. To overcome these problems, a novel algorithm called Multi-Branch Ferns (MBFerns), and R-package has been developed to build multi-branch ferns (multi-branch decision tree) and to generate key features from training dataset employing Naive Bayesian probabilistic model as classifier. The proposed algorithm performs well for general classification problems and extracting actionable knowledge from training data. The proposed method has been evaluated with best existing classification methods namely, Random Forest (RF), Support Vector Machine (SVM) and Artificial Neural Network (ANN) on medical benchmark data, available at https://archive.ics.uci.edu/ml/datasets/ such as Breast Cancer, Cryotherapy, Cardiotocography, Dermatology, Echocardiogram, EEG Eye State, Fertility, Haberman's Survival, Hepatitis, Indian Liver Patient, Mammographic Mass, Parkinsons, etc. Detailed investigation on proposed Multi-Branch Ferns (MBFerns) with respect to accuracy, time, space complexity and knowledge discovery has also been presented.