This study aims to predict the hardness of steel wire rods produced in the actual process by using statistical analysis and machine learning algorithms. The prediction model is built through five steps: operational data collection, key feature selection, experimental data collection, building a prediction model, and revising the model. We propose an innovative data collection methodology that mitigates the challenges posed by the wire rod's form and resolves the issue of mismatch between different measurement locations. The effects of features on hardness of wire rod are estimated with correlation analysis and explainable AI method. To ensure the model's robustness across varying process conditions, experimental data are collected via conducting a simulation test on key features that have a large effect. A hardness prediction model using experimental data, and it is validated and revised using operational data. Several statistical indices ensure that the prediction model has good prediction performance. The proposed method has contribution in that it provides a systematic procedure to predict the hardness of wire rods using operational data from the wire rod rolling process. It serve as an effective tool for predicting and optimizing the hardness of steel wire rods.