Mycobacterium tuberculosis (Mtb) β-carbonic anhydrases (β-CAs) are crucial enzymes responsible for regulating pH by catalyzing the conversion of CO2 to HCO3-, which is essential for its survival in acidic environments in the host. By inhibiting Mtb β-CAs, we can potentially discover new targets for anti-tuberculosis drugs with a different mechanism of action than existing FDA-approved drugs. This is crucial since Mtb has demonstrated the ability to develop different degrees of resistance to current drugs over time. This study employed machine learning-assisted quantitative structural activity relationship (ML-QSAR) models utilizing PubChem fingerprints, substructure fingerprints, and 1D 2D molecular descriptors to decipher the structural insights underlying the Mtb β-CA inhibition mechanism among 267 molecules. The final models, based on a random forest (RF) ML algorithm, demonstrated robustness with correlation coefficients of 0.931, 0.9227, and 0.9447, respectively. The final predictive models were further developed as a user-friendly web application, Mtb-CA-pred (https://mtb-ca-pred.streamlit.app/), which was further used to screen an anti-TB compound library of 11,800 molecules. We obtained two lead molecules, F0804-1219 and F1092-1799, from the virtual screening study, which were further subjected to a mechanistic systems biology framework to elucidate their inhibition mechanism through different biological pathways against Mtb β-CAs. Experimental validation via the minimum duration for killing (MDK) assay confirmed the bactericidal effects of the two identified compounds against Mycobacterium marinum biofilms, aligning computational predictions with experimental outcomes in drug discovery. These findings underscore the efficacy of the identified compounds as potent anti-TB agents, bridging computational and experimental approaches in anti-TB drug development.