Aluminum alloy is an important aerospace equipment material, and its element content is an important factor determining the quality and performance of aluminum alloy materials. The Mn is an important element in aluminum alloy, which can stop the recrystallization process of aluminum alloy and increase the recrystallization temperature. Quantitative determination of aluminum alloy composition is an important part of on-line detection of alloy composition. The signal fluctuation (laser energy fluctuation, plasma instability, sample inhomogeneity, etc.) and self-absorption effect influence the determination of trace elements in aluminum alloys by laser-induced breakdown spectroscopy (LIBS). In order to eliminate the bias caused by the self-absorption effect and signal fluctuation, a new method for detecting alloy content using LIBS technology combined with the LASSO-LSSVM machine learning method is proposed. The Least Absolute Shrinkage and Selection Operator (LASSO) model is used to select the spectral eigenvectors, reducing the dimension of the spectral data to match the training samples, reducing the risk of overfitting, and effectively extracting the most important features that characterize LIBS spectra. The Least squares support vector machine regression (LSSVM) model is used to train the characteristic spectra selected by LASSO. Compared with the internal standard method and partial least squares regression (PLSR), the analysis results show that the model accuracy and accuracy of LASSO-LSSVM were improved. The Mn element regression curves correlation coefficient (R2) of Mn element regression curve increased from 74.62% to 99.29%. The mean relative error (ARE) decreased from 22.38% to 3.56%, the root mean square error (RMSEC) of the training set decreased from 0.66 wt% to 0.040 wt%, and the root mean square error (RMSEP) of the test set decreased from 0.58 wt% to 0.042 wt%. The LASSO-LSSVM regression model is suitable for complex and high-dimensional spectral data with high uncertainty, and can greatly reduce input spectral datas dimension and redundant information. Therefore, the model reduces the overfitting problem of LSSVM. The results show that LIBS technology and the LASSO-LSSVM regression model can effectively improve the quantitative analysis performance of aluminum alloy materials by LIBS technology, which is a simple, reliable and high-precision method to detect alloy content.