This study proposes a geological disaster prediction model based on meteorological characteristics to address the frequent occurrence of geological disaster in rural mountainous areas. This model focuses on geological disaster significantly affected by meteorological conditions, including five types of geological disaster: mudslides, landslides, collapses, flash floods, mudslides, and landslides. Based on geological disaster data and meteorological observation data from 2011 to 2022, a 1:1 negative sample of geological disaster was constructed, and the Spearman correlation coefficient was used to screen out the 15 meteorological characteristic factors with the highest correlation with geological disaster. In terms of model construction, mainstream machine learning algorithms such as XGBoost, LightGBM, CatBoost, and RF were used to compare their accuracy and root mean square error on geological disaster samples. After comprehensive evaluation, the LightGBM algorithm performs the best in geological disaster prediction on this dataset. At the same time, an importance analysis of meteorological features was conducted based on this algorithm, removing meteorological feature factors with high Spearman correlation but little impact on geological disaster prediction, and optimizing the geological disaster prediction model based on LightGBM algorithm. Finally, the ROC curve and confusion matrix were used to evaluate the prediction model, and the AUC value of the model reached 0.99, an increase of 3% compared to before optimization; The accuracy of the model reached 0.98, an increase of 2% compared to before optimization, significantly improving the effectiveness of geological disaster prediction.