A total of 75,640 test-day milk yield records of 248 Holstein cows in the first (124 heads), second (75 heads), and third lactation (49 heads) were used as material in the study. All data used in this study were obtained from the database of the Afikim herd management software used on a private dairy farm. To predict 305-day adjusted milk yields (MY305) using some partial milk yield parameters, ALM (Automatic Linear Modeling), C&RT (Classification and Regression Tree), CHAID (Chi-square Automatic Interaction Detector), RF (Random Forest), MARS (Multiple Adaptive Regression Splines), Bagging MARS (Bootstrap Aggregating Multiple Adaptive Regression Splines), and BRNN (Bayesian Regularized Neural Network) data mining algorithms were used with group five-fold cross-validation. When all algorithms are compared in terms of 15 different prediction performance measures, the most successful algorithms are MARS (R2Adj = 0.844, RRMSE = 6.530 and MAPE = 5.182), Bagging MARS (R2Adj = 0.840, RRMSE = 6.547 and MAPE = 5.103), while C&RT (R2Adj = 0.828, RRMSE = 7.028 and MAPE = 5.542) is the most efficient tree-based algorithm. When the model evaluation criteria, including systematic bias and limits of agreement (LoA) among prediction performance measures, were examined together, the prediction success of the data mining algorithms was determined as MARS, Bagging MARS, C&RT, ALM, BRNN, CHAID, and RF, respectively. As a result, it can be stated that 75-day partial milk yield totals before peak milk yield is an important time period and an indirect selection criterion in determining 305-day milk yield. Additionally, it can help producers evaluate the impact of past milk yields on future cow productivity and predict overall herd performance, thereby facilitating timely and informed decision-making.