Advancing Predictive Geohazard Assessment for Underground Mines in the Nigerian Basement Complex Using Microseismic Monitoring and Machine Learning

Geohazard assessment in underground mines is conventionally reactive, responding to observed instability rather than predicting hazardous events before they occur. This dissertation advances the science of predictive geohazard assessment for underground mines in the Nigerian Basement Complex by developing and validating an integrated microseismic monitoring and machine learning prediction system. A network of triaxial geophones was installed at three underground gold and gemstone mines in Kaduna, Zamfara, and Ondo states, and 22 months of continuous microseismic data were recorded. A database of 18,400 seismic events was constructed, and source parameters including moment magnitude, b-value time series, seismic energy, and P-to-S wave amplitude ratio were extracted. Labelled training data were generated by correlating seismic event clusters with documented geohazard occurrences including roof falls, rib failures, and pillar cracking. Five machine learning classifiers were trained and tested using cross-validation, with a random forest ensemble model achieving the highest precision-recall performance for geohazard prediction 24 hours in advance. The system achieves a true positive rate of 79 percent and false positive rate of 11 percent in blind test data, representing a material improvement over human-only interpretation of microseismic trends. An original geohazard early warning scoring algorithm is derived from the model outputs and designed for real-time implementation. The dissertation delivers specifications for a commercially deployable early warning system and a geohazard data standard for Nigerian underground mines. Keywords: microseismic monitoring, geohazard prediction, machine learning, Nigerian Basement Complex, underground mining.