Designing a Risk Prediction Model based on a Deep Learning Algorithm with a Hybrid Approach

Risk management and forecasting is a constantly changing process. Constant evolution is inevitable, because the long-term performance of risk management cannot keep pace with recent developments or accurately predict emerging crises. Therefore, it is important to monitor and accept changes caused by structural failures in the risk management process. Adopting these changes requires redefining risk management components and tools. Traditionally, empirical research in finance has focused heavily on statistical inference. The purpose of this research is to use deep learning algorithm in order to provide a model for predicting the financial risk of companies. Therefore, it is developmental-applicative in terms of purpose. Considering that it has examined the problem of risk prediction at a general level and in a non-linear way, it is a holistic research in terms of paradigm.In terms of information gathering method, it is library research based on literature and theoretical background. Also, in terms of the research approach, it is exploratory (quantitative-qualitative). The desired method in qualitative analysis is metacomposite. The statistical community of the qualitative section includes scientific research articles and the statistical sample was determined to be 16 documents. In the quantitative part of the statistical society, there are companies active in the capital market of Iran. The statistical sample based on the systematic target method includes 199 active companies in the stock market between 2013 and 2014. The standard framework used to use the deep learning method is the TensorFlow program, which is a free and open-source library for "data flow programming" for machine learning and deep learning. MAXQDA software was used to code and analyze the content of the articles.The results showed that the error values of the training models in the deep learning approach in all cases of lasso regression, ridge regression, artificial neural network and random forest regression are less than 0.05 and the best method for machine learning is to use the mixed method of ridge regression and It is an artificial neural network.