Neural Network-Based Deep Encoding for Mixed-Attribute Data Classification

摘要: This paper proposes a neural network-based deep encoding (DE) method for the mixed-attribute data classification. DE first uses existing one-hot (OE) to encode discrete-attribute data. Second, trains an improved network classify OE-attribute corresponding The loss function of not only includes training error but also considers uncertainty hidden-layer output matrix (i.e., DE-attribute data), where is calculated with re-substitution entropy. Third, classification task conducted based on combination previous continuous-attribute and transformed Finally, we compare OE by support vector machine (SVM) (DNN) 4 KEEL sets. experimental results demonstrate feasibility effectiveness show that can help SVM DNN obtain better accuracies than traditional method.