I want to build a robust machine-learning pipeline that can reliably predict when a user is likely to log on again. The core need is a production-ready model—deep-learning is welcome where it adds value—that captures behavioural signals and translates them into accurate logon-probability scores. You will work with the raw system logs I can provide (time-stamped events, account metadata and any other fields you advise extracting). Starting from exploratory data analysis, we will move through feature engineering, model selection, hyper-parameter tuning and final evaluation. I am especially interested in interpretable insights alongside raw accuracy, so attention to explainability techniques such as SHAP or LIME is appreciated. Deliverables • Clean, well-commented Python (or R) code for data prep, training and inference • A trained model saved in a reusable format (Pickle, ONNX, or similar) • A concise report summarising methodology, feature importance and validation metrics • Step-by-step instructions for recreating the results on my environment Acceptance criteria: A minimum AUC of 0.80 on hold-out data and clear documentation that lets another engineer reproduce the workflow without guesswork. If this sounds like your field, let’s discuss the data snapshot and the milestones so you can get started right away.