This study describes a method for constructing a machine learning model to correct bias in a chemical transport model. When the neural network model was trained with data periods ranging from one to three years, it was observed that extending the training period enhanced the accuracy of the bias correction. This improvement is attributed to the increased variety of meteorological characteristics and air pollution patterns incorporated into the model, resulting in a greater versatility. A comparison of bias correction accuracy between machine learning models and a simple linear regression model revealed that all the machine learning models performed better. Specifically, for data with observed values of 80 ppb or more, the linear regression model exhibited greater error and a more pronounced tendency toward underestimation compared to the machine learning models. To address the oversight of high concentrations, a weighting factor was applied to the data with observed values of 80 ppb or more and calculated values more than 30% lower than the observed values, and the neural network model was trained. Consequently, the proportion of data with observed values of 80 ppb or more that fell within a 30% error margin increased to 90%.