The data-driven fault detector is a tool to detect failure occurrences in dynamical systems. In particular, using historical data collected from a system operating within its nominal (i.e., non-faulty) behavior, the tool creates a nonlinear dynamical model, instead of a simple linear model, of the system via Machine Learning-based algorithms. Then, it sets up a Kalman filter whose state consists of the model parameters and runs a monitoring procedure to detect if unexpected behaviors occur.