Multi-Task LSTM-Attention with Adaptive Isolation Forest for Intelligent Project Implementation Monitoring

Abstract Manual oversight of funded research projects does not scale. Progress delays, budget irregularities, and superficial reportingoften go undetected until final acceptance, while existing detection methods struggle to distinguish legitimate scheduleadjustments from genuine anomalies. We present the Intelligent Project Monitoring System (IPMS), which couples a featuredecoupled multi-task LSTM-Attention network with an Adaptive Isolation Forest. The LSTM-Attention component modelsproject workflows through finite state machines and predicts milestone deviations; the Adaptive Isolation Forest flags anomaliesafter a legitimate deviation filter screens out justifiable variances that would otherwise inflate false-positive counts. A multi-headattention module separately tracks how execution performance evolves over the project lifecycle, and all components updateonline when distribution shifts are detected. On a real-world dataset with approximately 7% confirmed anomalies, IPMSachieved an AUC of 0.924 and a false-positive rate of 4.2%, a 76.9% reduction relative to standard Isolation Forest. Milestonedeviation prediction reached an MAE of 1.85 days, 34.2% lower than standard LSTM. Execution profiling scored an MAE of0.082. Removing the deviation filter alone degraded F1 by 12.3%, and removing multi-scale fusion increased the miss rate forlong-duration stalls by 23%.