A Financial Distress Prediction Model Based on Orthogonal Projection Reduction by Affinity and Probabilistic Neural Network

摘要

In order to improve the prediction accuracy and efficiency of the financial distress prediction (FDP) model, this research proposed a model combining orthogonal projection reduction by affinity (OPRA) and probabilistic neural network (PNN).A novel feature extraction approach -OPRA approach was to select the key financial indicators and reduce dimensions by introducing Locally Linear Embedding (LLE) into the traditional method of Principal Component Analysis (PCA). The OPRA approach not only kept the reservations of financial indicators' data flow-shaped structure during high-dimensional to lowdimensional mapping but also got better orthogonal projection.Empirical results showed that OPRA could obtain a better dimension reduction performance in FDP and a FDP model based on OPRA and PNN could produce a better prediction accuracy and computational efficiency in FDP than PNN approach.