Objective: The aim of this study is to investigate whether a reinforcement learning (RL) algorithm is effective to improve the accuracy of the safety culture categorization which the author's previous study tried to solve with a convolutional neural net classifier.
Background: The RL algorithm using the optimal Bellman equation is nowadays popularly applied to many markov decision process situations.
Method: An asynchronous advantage actor critic (A3C) neural net was applied to learn the safety culture survey data collected from a nuclear power industry for safety culture level categorization.
Results: The RL algorithm applied to the randomly selected validation data resulted in 97.63% accuracy compared to 96.2% which showed in the previous study.
Conclusion: The safety culture level categorization using the A3C neural net is more stable and accurate than the previous convolution neural net classifier.
Application: The safety culture level classifier with the RL neural net learning massive survey data might be useful in place of expert interviewers for safety culture evaluation.