SODM-Based Selective Neural Networks Ensemble for Regional Innovation Capability Evaluation

At present, the single evaluation method is adopted usually in regional innovation capability evaluation. This paper combines neural networks ensemble and self-organization data mining (SODM), and proposes SODM-based selective neural networks ensemble (SSNNE) evaluation model. SSNNE includes two stages. Firstly, it selects the key indexes from the evaluation system that affect the regional innovation capability with SODM; secondly, it trains some neural network evaluation models, conducts ensemble selection with SODM and gets the final evaluation result. Compared with seven commonly-used single evaluation methods such as principal components analysis, Topsis and grey relational analysis, the empirical results show that SSNNE can get better evaluation results.