Real-time prediction of project duration deviation using CNN-LSTM-FIS

Purpose – This study aims to develop a real-time system for monitoring deviations in construction project durations, enabling project managers to detect schedule delays early and take proactive measures to mitigate risks. Design/methodology/approach – A hybrid predictive framework combining convolutional neural networks (CNN), long short-term memory (LSTM) and a fuzzy inference systems (FIS) is proposed. The CNN models classify construction activities from daily site images, while the LSTM networks predict activity durations based on time-series data extracted from metadata and daily reports. Deviations are identified by comparing predicted and planned completion times. The FIS is used to incorporate uncertainty and interpret deviation predictions and classify deviation predictions severity in real time. Findings – The model was validated using data from a water reservoir construction project in Iran. The CNN achieved 90% classification accuracy, and LSTM reported a Mean Absolute Error of 0.2. The integrated system achieved a mean absolute percentage error of 14.65% for project deviation, indicating good predictive performance. Research limitations/implications – The model was evaluated on a single infrastructure project. Future studies should assess its generalizability across diverse project types, scales and conditions. Practical implications – The system supports proactive construction management by integrating real-time visual and contextual data. It reduces reliance on manual progress tracking and facilitates early corrective actions to avoid cost and schedule overruns. Originality/value – This framework uniquely combines sequential forecasting, image classification and fuzzy logic to address limitations of previous models and advance intelligent construction monitoring systems.