Use of artificial intelligence for risk forecasting and optimization of construction project implementation processes

Abstract

Risk forecasting and the reduction of deviations in schedule and cost remain a weak point in the delivery of construction projects, where a noticeable share of facilities is completed with overruns of planned indicators. The gap between deterministic schedule-and-estimate calculation and the probabilistic nature of deviations produces a systematic underestimation of reserves. The aim of the study is to quantify the contribution of a machine-learning predictive layer, embedded into planning and operational management, to the reduction of project deviations. On an author's simulation model of a monolithic-frame residential facility with a planned cost of 2,847.3 million rubles and a duration of 26 months, a Monte-Carlo calculation (10,000 iterations) was performed over eight risk factors, supplemented by a comparison of four algorithms for forecasting the final cost and by a variance decomposition of the factor contribution. The baseline scenario yields an average estimate overrun of 6,4% and a schedule overrun of 2,2 months, with an 80th percentile of 10,3% and 30.4 months. Connecting the predictive layer lowers the average overrun to 3,4% and the average delay to 1,1 months, cutting the expected deviations roughly by half. The integral indicator shows the highest sensitivity to the financing-rate risk and to supply disruption, which together form more than 40% of the variance of the final cost. The distribution of the factor contribution shows that the effect of predictive algorithms is achieved by re-tuning already applied procedures of schedule and cost control, without a separate management block.