The optimization of the financing structure (debt-equity) continues to be a central problem in corporate finance due to its simultaneous impact on risk, weighted average cost of capital (WACC) and profitability. In recent years, artificial intelligence (AI)—in particular, machine learning (ML) and explainable approaches (XAI)—has expanded the ability to estimate target leverage, anticipate financing decisions, and model nonlinear relationships with heterogeneity across firms and over time. This paper develops an empirical approach with panel data that integrates (I) ML to approximate "optimal" capital structure (via prediction of cost of capital and/or target leverage) and (II) panel econometrics to assess the association between closeness to the financial target and profitability (ROA/ROE). Based on recent evidence, it is observed that ML models outperform linear specifications in the prediction of leverage and its determinants, increasing out-of-sample performance; in addition, the interpretability based on SHAP values facilitates the traceability of financial drivers. In parallel, panel studies document that capital structure significantly affects profitability (with industry-dependent outcomes) and that firms gradually adjust toward target debt levels. Implications for financial management, risk control, and model governance are discussed.