Deception detection is a big deal in areas like national security, privacy, and courtrooms. It's tough to tell truth from lies because there are so many factors involved. Traditional lie detectors, or polygraphs, have been around for a while but have issues. They are controversial because of scientific, ethical, and practical problems.

Now, machine learning is making things easier. But there's a catch. Most datasets used for training these systems are small, use only one type of data, and are based on non-Indian populations. This is a problem because people from different backgrounds might show different signs when they lie.

To fix this, researchers created CogniModal-D, a new dataset. It includes data from over 100 people and uses seven different types of measurements. These include brain activity, heart rate, eye movements, and more. The data was collected during social tasks and mock crime interrogations. This makes it more realistic and useful.

The study found that using multiple types of data (multimodal) is better than using just one. This is because different people show different signs when they lie. For example, some people might change their heart rate, while others might change their eye movements. By looking at multiple signs, the system can be more accurate.

The researchers also found that some types of data are better than others. Behavioral data, like audio, video, eye gaze, and skin response, were more reliable than neurophysiological data, like brain activity and heart rate.

This study shows that using multiple types of data can make deception detection more accurate. It also highlights the need for more diverse datasets. This is important for developing better, more reliable deception detection systems in the future. It's a step forward in making sure we can trust the technology we use to catch liars.

Catching Liars: The Science Behind Spotting Lies

Actions