 The study evaluates and compares three popular neural network models for landslide susceptibility mapping in Qingchuan County, Sichuan Province, China. ANN and 1DCNN models yielded consistent results while RNN outperformed the others but was complex. Experiments suggest that small batch size values with TAN activation function and SGD optimizer improved the performance of neural network models for LSM. This article was authored by Yang Yingyi, Wang Changzhong, Xu Weishu, and others.