To forecast the endpoint of the Basic Oxygen Furnace (BOF) process in steelmaking, we have employed deep learning techniques. However, our project faces limitations due to insufficient data, leaving key influencing factors undisclosed at the process's conclusion. The BOF process is intricate and multi-targeted, primarily managed manually by operators. It involves converting a blend of pig iron and recycled scrap into low-carbon steel. Our strategy involves deploying a joint neural network and comparing it against a static model to evaluate whether incorporating sequential data enhances predictive precision. Trained deep learning models exhibit proficiency in accurately predicting temperature, carbon, and phosphorus within predefined limits. We did SHAP analysis for finding the influential factors for target variables. Leveraging a comprehensive dataset, we conducted predictions on these target variables. One model relies solely on static data, while the other is a joint model integrating static and sparse sequential data. Surprisingly, the accuracy of the static model surpasses the joint model, with R2 scores of 0.92 for phosphorous, 0.79 for temperature, and 0.71 for carbon compared to lower R2 scores for the joint model, indicating that richer data can indeed enhance predictions in the BOF process.