Learning Management System Predictive Dropout Detection

Create a comprehensive predictive model using Python's TensorFlow and Keras to identify students at high risk of academic dropout. The model should integrate multiple data sources including LMS interaction logs, grade history, attendance records, and demographic information. Implement a multi-layer neural network with at least 3 hidden layers, use cross-validation techniques, and generate a probability score for potential dropout risk. Include a modular design allowing easy retraining and feature updates.