Abstract: In the traditional methods of meteorological risk early warning and forecasting, the vulnerability factors of disaster bearing bodies are ignored when classifying the meteorological risk level, and the meteorological risk prediction level is relatively high, which leads to the high air report rate in high-level risk areas. Based on this, a method of early warning and forecasting of meteorological risk of landslide and collapse geological disasters based on machine learning is proposed. By using the information quantity method, the influence degree of meteorological factors is analyzed, and the coordinate point, rainfall and prone level are selected as input nodes of machine learning artificial neural network to judge whether geological disaster occurs; for the area of ground damage, the meteorological cause sub index is calculated according to the influence degree. Combined with the potential degree of geological disaster and vulnerability of disaster bearing body, the meteorological risk warning index is determined, divide the warning and forecast level, and complete the forecast of geological disaster meteorological risk. The experimental results show that the proposed method can effectively reduce the three-level forecast air report rate and the fourth level air alarm rate, and improve the precision of the early warning forecast.