The ADMIXTURE algorithm has been enhanced to improve individual ancestry estimation. These enhancements include cross-validation for determining the number of ancestral populations, supervised learning for more accurate ancestry estimates, penalized estimation to promote model parsimony, and parallel processing for faster analysis. The algorithm uses a parametric model to estimate ancestry fractions and population allele frequencies. Cross-validation helps identify the best number of ancestral populations by evaluating prediction error. Supervised learning uses known ancestral populations to improve accuracy. Penalized estimation reduces overfitting by discouraging small admixture coefficients. Parallel processing speeds up analysis using multiple processors. These improvements make ADMIXTURE more accurate, efficient, and versatile for ancestry estimation. The algorithm is available for Linux and Mac OS X, with C++ programming language. It is freely available as binaries, with proprietary source code. The enhancements allow ADMIXTURE to replace STRUCTURE in most applications, especially with large datasets. The algorithm is suitable for both exploratory and focused studies of genetic ancestry. The results show that supervised analysis provides more accurate ancestry estimates and runs faster than unsupervised analysis. Penalized estimation reduces bias in ancestry estimates, particularly for small datasets or closely related populations. The algorithm is widely applicable in genetic research and population genetics.The ADMIXTURE algorithm has been enhanced to improve individual ancestry estimation. These enhancements include cross-validation for determining the number of ancestral populations, supervised learning for more accurate ancestry estimates, penalized estimation to promote model parsimony, and parallel processing for faster analysis. The algorithm uses a parametric model to estimate ancestry fractions and population allele frequencies. Cross-validation helps identify the best number of ancestral populations by evaluating prediction error. Supervised learning uses known ancestral populations to improve accuracy. Penalized estimation reduces overfitting by discouraging small admixture coefficients. Parallel processing speeds up analysis using multiple processors. These improvements make ADMIXTURE more accurate, efficient, and versatile for ancestry estimation. The algorithm is available for Linux and Mac OS X, with C++ programming language. It is freely available as binaries, with proprietary source code. The enhancements allow ADMIXTURE to replace STRUCTURE in most applications, especially with large datasets. The algorithm is suitable for both exploratory and focused studies of genetic ancestry. The results show that supervised analysis provides more accurate ancestry estimates and runs faster than unsupervised analysis. Penalized estimation reduces bias in ancestry estimates, particularly for small datasets or closely related populations. The algorithm is widely applicable in genetic research and population genetics.