This paper presents a novel approach to constructing an analytic holographic model of Quantum Chromodynamics (QCD) using machine learning techniques. The authors leverage lattice QCD results for the equation of state (EOS) and baryon number susceptibility at zero chemical potential to build the Einstein-Maxwell-Dilaton (EMD) framework for pure gluon, 2-flavor, and 2+1-flavor systems. The EMD framework is a non-perturbative method that describes QCD matter under extreme conditions, such as those found in heavy ion collisions and neutron stars. The authors use a deep neural network to regression analyze the lattice QCD data, optimizing the parameters of the EMD model to match the lattice results. The model predicts the critical endpoint (CEP) in the 2+1-flavor system at $(T^c = 0.094 \text{GeV}, \mu_B = 0.74 \text{GeV})$, which aligns well with results from other non-perturbative models like the functional renormalization group (FRG) and the Polyakov-Nambu-Jona-Lasinio (PNJL) model. The paper also compares the predicted dilaton potentials with those from the extended DeWolfe-Gubser-Rosen (DGR) model, showing good agreement. This validation confirms the robustness of the EMD framework in describing QCD matter. The phase diagram in the $(T, \mu_B)$ plane is constructed, revealing that the phase transition is crossover at small chemical potentials and first order at large chemical potentials for both 2-flavor and 2+1-flavor systems. The study highlights the potential of machine learning in constructing realistic holographic models of QCD, providing insights into the phase structure of QCD matter and aiding in the search for the CEP in heavy ion collisions.This paper presents a novel approach to constructing an analytic holographic model of Quantum Chromodynamics (QCD) using machine learning techniques. The authors leverage lattice QCD results for the equation of state (EOS) and baryon number susceptibility at zero chemical potential to build the Einstein-Maxwell-Dilaton (EMD) framework for pure gluon, 2-flavor, and 2+1-flavor systems. The EMD framework is a non-perturbative method that describes QCD matter under extreme conditions, such as those found in heavy ion collisions and neutron stars. The authors use a deep neural network to regression analyze the lattice QCD data, optimizing the parameters of the EMD model to match the lattice results. The model predicts the critical endpoint (CEP) in the 2+1-flavor system at $(T^c = 0.094 \text{GeV}, \mu_B = 0.74 \text{GeV})$, which aligns well with results from other non-perturbative models like the functional renormalization group (FRG) and the Polyakov-Nambu-Jona-Lasinio (PNJL) model. The paper also compares the predicted dilaton potentials with those from the extended DeWolfe-Gubser-Rosen (DGR) model, showing good agreement. This validation confirms the robustness of the EMD framework in describing QCD matter. The phase diagram in the $(T, \mu_B)$ plane is constructed, revealing that the phase transition is crossover at small chemical potentials and first order at large chemical potentials for both 2-flavor and 2+1-flavor systems. The study highlights the potential of machine learning in constructing realistic holographic models of QCD, providing insights into the phase structure of QCD matter and aiding in the search for the CEP in heavy ion collisions.