Global Data Science Forum

 View Only

Risk Modelling

By Moloy De posted Thu December 31, 2020 08:46 PM

  
We need to perform effort estimation prior to a project. Many a times managers are asked to reduce the estimated effort in order to fit the budget. But that should come with an increased risk in completing the project in time. I was asked to model the risk against the reduction in effort hours so that one can readily communicate it based on data.

Following were the basic guidelines for modelling the risk / confidence:
    1.    With 40% of the estimated hours the confidence for completing the project in time will be minimum
    2.    With 225% of the estimated hours the confidence for completing the project in time will be maximum
    3.   With 100% of the estimated hours the confidence for completing the project in time will be 75%

Beta Distribution is a proper candidate to model this risk / confidence as the distribution support can be manipulated to be any positive range and the shape parameters (alpha only) can be adjusted to control the third quartile. Other skewed distributions like Gamma or Lognormal do not have Bounded support and cannot be used. However Weibull Distribution for which closed form expressions of Quantiles exist is found to be not workable with.

It is impossible to solve equations as below to have explicit expressions for Alpha and Beta, the only parameters of the Beta Distribution. So excel Beta-Inverse function is used to solve them numerically.


Using the Macro, client was capable of finding the change in confidence levels when client is queried on a percentage variation in the project effort estimate.
The calculation macro was handed over to client in five weeks time and that helped the client to replace Oracle Crystal Report saving them the cost of the license.

QUESTION I : How the data was used to train the model?
QUESTION II :  How can one justify the guidelines provided?

#risk-modelling
0 comments
3 views

Permalink