This study quantitatively analyzes the major elemental content in Martian-like minerals using laser-induced breakdown spectroscopy combined with a feature fusion strategy. First, the spectral data are preprocessed, and the characteristic wavelength of each element is identified through qualitative analysis to determine the characteristic peak area (SK). Simultaneously, principal component analysis (PCA) and partial least squares (PLS) are combined to perform two-level feature extraction of the spectral data, yielding feature extraction variables. Finally, this variable and SK are fused and input into a genetic algorithm-optimized support vector machine regression (GA-SVR) model. Upon comparison with models established using SK, PLS, PCA alone, and the partial least squares regression (PLSR), the feature fusion (FF) -GA-SVR model shows root-mean-square errors (RMSEs) that are, on average, lower by 33.9%, 32.8%, 22.9%, and 24.1%, respectively. In addition, the model achieves an average coefficient of determination (R²) of 0.961, demonstrating its robust predictive performance. This highlights its potential as a foundational approach for rapid and accurate analyses of unknown ore sample compositions.