I'm training a machine learning model on a dataset related to pump failure prediction. However, I'm noticing that the accuracy (and other metrics like ROC AUC) change every time I run the notebook, even though I haven't changed anything in the code or data. I suspect this might be due to randomness in train/test splitting, SMOTE resampling, or model initialization, even overfitting? But I'm not sure how to control it properly. my code reference is based on this: https://github.com/Fazil-Ahamed-M/P...ob/main/Notebook_Predictive_Maintenance.ipynb Continue reading...
