In this study, we compared a simple GIS thresholding approach with a random forest machine-learning model for mapping landslides using multi-temporal LiDAR DEMs in a 150 km squared area of Baltimore County, Maryland. We derived key terrain indicators: elevation change, slope, wetness index, curvature and stream power from successive DEM surveys. The GIS method flagged potential slides using fixed cutoff values and filtered out small artifacts, while the machine learning model learned to distinguish true slides from other features based on expert-mapped examples. Although the GIS rules captured most real slides, they also produced many false alarms on urban cut slopes. By contrast, the random forest achieved a more balanced detection rate and retained its accuracy when applied to new survey data without additional training. Our results show that integrating LiDAR-derived terrain metrics with machine learning delivers more accurate and consistent landslide inventories over time, supporting scalable and reliable hazard monitoring across varied landscapes.