Understanding in situ mechanical rock properties is critical for wellbore stability, hydraulic fracturing, reservoir characterisation and remediation/abandonment operations. These properties are determined using specialised wireline log data and performing well and drill core testing to calibrate and constrain the log-derived models. However, a significant number of wells lack the required logs and tests, making accurate mechanical characterisation a significant challenge. This study presents a novel approach for the prediction of mechanical properties in these data-limited well scenarios. A specific application for this is identifying ‘weak’ points in a well to guide cement-plug pressure testing during well abandonment operations. In the area of interest for this work (Surat Basin), the most common dataset for development wells is a ‘triple-combo’ wireline logging suite. Curves acquired using this logging suite include gamma-ray, bulk density, neutron porosity and resistivity data. The proposed methodology leverages machine-learning and rock-physics relationships to predict mechanical properties in data-limited wells. Key wells were identified as having the required wireline log data, well test data and core testing to characterise and model in situ mechanical properties. Data from these wells were then used to train and test a machine-learning model and constrain rock-physics relationships allowing for characterisation of mechanical properties in data-limited wells. This work addresses a pressing industry need and is particularly relevant to operations with high well density, reducing costs and improving well integrity and regulatory compliance. Also, it highlights the potential of machine-learning and data integration in improving our understanding of subsurface rock properties in data-constrained settings.