This study proposes an affine term structure model (ATSM), which incorporates 129 factors with their interactions (five standard yield factors plus 124 macro-financial factors), and implements machine learning with no-arbitrage conditions. First, our empirical model fits yields and predicts future excess returns with fast computation - the larger the number of macro-financial factors, the better the performance. Through Lasso regression combined with principal component analysis, we illustrate how machine learning helps identify 23 macro-financial variables to predict bond return. The results yield specific economic implications such that yield curve dynamics explicitly covary with housing permits, short-term rates, stock prices, labor market, and inflation. Our data augmentation facilitates machine learning and enhances model performance. In sum, our ATSM mitigates the long-standing challenges of affine models: the small sample size, computational complexity to process numerous macro-financial factors, and the use of latent variables that makes economic interpretations difficult.