In this study, machine learning techniques were applied to model and optimize the electrical conductivity of aluminum alloys through predictive analysis and compositional design. Based on 371 sets of actual industrial production data, four regression models, support vector regression (SVR), decision tree (DT), multilayer perceptron (MLP), and XGBoost, were developed. The results demonstrate that the Extreme Gradient Boosting (XGBoost) model outperforms the other models in terms of mean squared error (MSE) and coefficient of determination (R2). Furthermore, by integrating the Bayesian optimization algorithm, the inverse design of aluminum alloy conductivity was within a defined compositional range. The optimized component scheme, experimentally verified, achieved a conductivity of 30.4% International Annealed Copper Standard (% IACS), surpassing the highest conductivity of 28.08% IACS from the original dataset, thereby validating the effectiveness of this method. Additionally, by combining thermodynamic calculations (using Pandat® software) with microstructure analysis, the study revealed the mechanisms by which element solid solution states and precipitation behaviors affect conductivity performance. The findings indicate that the low solubility of elements like Zn and Cu reduces electron scattering, enhancing the feasibility of the “low-solute and high-precipitation” design strategy. This study has developed an integrated framework for designing the conductivity of aluminum alloys, incorporating machine learning, optimization design, and microstructure analysis, providing new insights into the intelligent development of high-conductivity aluminum alloys.