A passive under the hood captcha system.
An alternate for the current captcha that doesn't need the user to interact and solve to prove itself human. This solution runs on the web and determines if it's a bot.
In addition to the system we release a dataset containing the mouse movements of a bot and a human to help train a model which can predict the user using the website is a crawling bot or a human. Please note that the dataset is standard scaled with a mean of 0.1705401898887262 and variance of 24.22827724591758 . The dataset is present in the Mouse_Movement_Dataset folder above . A plot of the mouse movements of a bot and a human to help realize the difference between them is given below:
The first image shows the mouse movements of a human and the second one shows that of a bot.
We also release a trained model with weights based on Long Short Term Memory(LSTM) Networks stored in the Trained_Model_and_Weights folder.
Please cite the repo and mail about the same to (pradyumna.gupta22@gmail.com) if using for publication purposes.