This dissertation presents an approach to facilitate the development of Lennard-Jones parameters for specific atom pairs, termed off-diagonal Lennard-Jones parameters, for use in empirical force fields. The study integrates molecular dynamics simulations, quantum mechanics, and machine learning to optimize Lennard-Jones parameters, enabling accurate modeling of non-covalent interactions and thermodynamic properties. A Molecular Interaction Rule (MIR) workflow was developed to automate potential energy scans for diverse molecular dimers, leveraging internal coordinate systems for systematic interaction studies. Key thermodynamic observables—enthalpy of vaporization, sublimation, and mixing, were incorporated into a ML-driven optimization workflow that may potentially enhance the predictive accuracy of binary liquids, including fuel mixtures. The research explores advanced optimization algorithms such as Monte Carlo, Bayesian Optimization, and Particle Swarm Optimization to balance van der Waals forces across functional groups through optimization of the Lennard-Jones parameters. Mixed solvent simulations were validated against experimental data, revealing critical insights into off-diagonal Lennard-Jones parameter tuning. The developed workflow and methodology will facilitate the development of empirical force fields for a wide range of uses including biology, drug design and fuel design. Additionally, the framework is presented for a ML architecture employing wave-transform kernels to capture three-dimensional molecular features and refine policy networks for parameter convergence.