Design of Hyperparameter Tuned Deep Reinforcement Learning Based Prediction Model for Financial Crisis

Financial crisis prediction (FCP) greatly contributed to the country's economic phenomenon. Precise estimation of the number and probability of failure indicated the strength and advancement of the national economy. Generally, different methods are planned for a potential FCP. In contrast, predictive accuracy, classifier efficiency, and data legality are not optimal for practical application. Reinforcement learning (RL) methods define how an agent can only study the best action policy in a sequential decision procedure through experience. The agent policy offers him some terminal and running rewards in a given situation. In online learning, the agent learns sequentially. This paper designs a novel Hyperparameter Tuned Deep Reinforcement Learning based Prediction Model for Financial Crisis (HPTDRL-PFC) technique. The goal of the HPTDRL-PFC technique lies in the identification of financial crises. To do so, the presented HPTDRL-PFC technique primarily normalizes the input financial data into a uniform format. For the prediction process, the Bayesian deep Q-Network (BDQN) technique is applied in this work. To improve the predictive ability of the BQDN technique, the artificial hummingbird optimization (AHBO) algorithm performs the hyperparameter tuning process. The performance validation of the HPTDRL-PFC technique is tested using a financial dataset. The comparative study reported the enhanced prediction outcomes of the HPTDRL-PFC technique over other models.