Unsupervised Anomaly Detection Method - Multivariate Time Series through LSTM Autoencoder

Job ID: 33055473

Budget: $250 – $750 CAD

We need to build a "Mouse Dynamics Based Bot Detection Model Using an Unsupervised Anomaly Detection Method - Multivariate Time Series through LSTM Autoencoder"

A set of data has been collected from users' mouse movement behavior on a website including (Session, Timestamp, Button, Event Type, State, x and y coordinates, Speed). Besides, there is a program that generates automated mouse movement on the designated website. Considering that, the main tasks in this project are as follows:

*Preprocess the data

*Extract couples of features from data including Acceleration, Mouse action duration/elapsed time, Movement speed in X and Y directions, Directional features (angle of movement)- The direction of movement at a given timestamp

*There probably exists some ‘lower’ threshold on the ‘deviation’ between repeat human sessions. There is a minimum distance between the sessions generated by a single human user. Through data analysis, we want to determine the minimum distance between two consecutive sessions.

*Train the LSTM

*Find the error

*Visualization in each step might be necessary

The project can be delivered in different steps, the code with explanation will be delivered.