Can Machine Learning models effectively predict project risks better than traditional Monte Carlo?

All answers to this question.

It depends entirely on the quality of your historical data. Monte Carlo is great because it’s transparent, but ML can identify non-linear relationships that humans (and simple stats) miss. For instance, an ML model might find that projects led by a specific department during Q3 always slip because of an undocumented resource conflict. We saw about a 12% improvement in our schedule accuracy after moving to a Random Forest model. But be warned: if your data is "dirty" or inconsistent across years, the model will give you very confident, very wrong answers.

   Answered 2024-06-24 by Deborah Young

Since you mentioned the "Random Forest" model, did you find it difficult to explain those results to the stakeholders who are used to simple bell curves?

   Answered 2024-06-27 by Joshua Taylor

• Joshua, that was our biggest hurdle. Stakeholders hate "black boxes." We had to use SHAP values to explain which features were driving the risk scores. It basically translates the ML logic back into "Human Speak." Once they saw that the model was flagging "unclear requirements" as the top risk factor, they started to trust the data-driven approach much more.

     Commented 2024-06-28 by Brian Harris

ML is great, but don't ditch Monte Carlo. We use ML to set the input parameters for our simulations, combining the best of both worlds for our executive reports.

   Answered 2024-06-30 by Karen White

• This hybrid approach Karen mentioned is exactly what we are looking into. Using ML as a "tuner" for existing models seems like the most logical first step.

     Commented 2024-07-01 by Barbara Martinez