Domestic and foreign studies have shown that aircraft arrive runway occupation time (AROT) is an important factor affecting airport runway capacity, and accurate prediction of runway occupation time is beneficial for more accurate evaluation of runway capacity. Due to the dynamic and complex nature of the landing process, a convolutional neural networks (CNN) focusing on data feature extraction was used to predict AROT, and the sparrow search algorithm (SSA) was used to optimize CNN-related parameters to overcome the problem of CNN easily falling into local optima. The data source used was the quick access recorder (QAR) records of aircraft, covering a total of 34 airports. Based on QAR data analysis, the SSA-CNN prediction model was constructed to analyze the factors affecting AROT. The analysis of QAR data showed that AROT was strongly correlated with taxiing distance, landing temperature, runway entrance speed, the number of rapid exit taxiways, and exit speed, while it had a relatively low correlation with factors such as aircraft weight, wind speed, wind direction, and exit taxiway angle. Based on the correlation of influencing factors, the mean square error (MSE) of the CNN prediction model was 18.35, while that of the optimized SSA-CNN prediction model was 17.31. The predicted results can provide reference for airport runway capacity evaluation.