Using Machine Learning and GA to Solve Time-Cost Trade-Off Problems

Existing genetic algorithms (GA) based systems for solving time-cost trade-off problems suffer from two limitations. First, these systems require the user to manually craft the time-cost curves for formulating the objective functions. Second, these systems only deal with linear time-cost relationships. To overcome these limitations, this paper presents a computer system called MLGAS (Machine Learning and Genetic Algorithms based System), which integrates a machine learning method with GA. A quadratic template is introduced to capture the nonlinearity of time-cost relationships. The machine learning method automatically generates the quadratic time-cost curves from historical data and also measures the credibility of each quadratic time-cost curve. The quadratic curves are then used to formulate the objective function that can be solved by the GA. Several improvements are made to enhance the capacity of GA to prevent premature convergence. Comparisons of MLGAS with an experienced project manager indicate that MLGAS generates better solutions to nonlinear time-cost trade-off problems.