A Robust, Resilience Machine Learning With a Risk Approach for Project Scheduling

This study proposes a novel Robust, Resilient, and Risk-Based approach in Machine Learning (3RML) that emphasizes the application of project scheduling for the first time. A robust stochastic LASSO regression model is proposed to predict project duration. This model seeks to enhance a traditional LASSO regression by minimizing the expected value and the Weighted Value at Risk (WVaR) of the Mean Absolute Deviation (MAD) while penalizing the regression coefficients. The 3R requirements, which prioritize robustness, resilience, and risk aversion, are integrated into the mathematical model to ensure flexibility and disaster consideration. A comparative analysis was carried out between the square root, logarithm, and mixed linear/square root models and the baseline model. The Robust, Resilience MAD with Risk-Averse (RRMADR) and R-squared values were computed. The square root regression model demonstrated a 36% enhancement compared with the primary model. The conservatism coefficient affects risk levels, where a 5% increase results in a 2% decrease in the RRMADR. Varying confidence levels influence the model. The penalty coefficient in the lasso regression affects RRMADR and R-squared. The resiliency coefficient impacts both the RRMADR and R-squared. Probability scenarios influence RRMADR but do not affect R-squared. The type of probability density influences the RRMADR but does not impact R-squared.