This study proposes the use of Machine learning (ML) alongside Raman spectroscopy (RS) to predict biomechanical properties for fracture-risk prediction. 118 human cadavers’ cortical femur bones were used to collect Raman spectra data as well as demographic features and structural parameters. Data was then processed to be split among four machine learning algorithms for training, Support Vector Regression (SVR), Extreme Gradient Boosting (XGB), Extra Trees Regression (ETR), and an ensemble of all three, with each being chosen for their ability to find complex non-linear relationships. Data was split up with 80% being used for training and 20% used for testing. Additional data derived from RS was added variably to test the effect of RS integration on ML prediction. Crack-initiation toughness (K-init) and energy-to-fracture (J-integral) were selected as the measurements that would be compared for accuracy. The study found that in both predictions of K-init and J-integral, the ensemble model outperformed all of the individual models, with higher R squared values and lower error metrics, and significant improvements in accuracy with RS derived data being integrated into the algorithm. The discussion finds that there is significant value in further research into RS data integration, but also addresses issues with generalizability due to heterogeneity of the study population. In future studies, independent data sets should be used, alongside more varied data.