Machine learning anxiety assessment: Decoding stress through colouring

Scientists have trained artificial intelligence to look for signs of stress in how people colour. This new tool analyses digitised mandala drawings to spot hidden visual clues. By developing a machine learning anxiety assessment, researchers offer a fresh way to support mental health evaluations. It is a brilliant example of how modern computing can assist traditional psychology.

The Problem: A need for a machine learning anxiety assessment

Doctors normally measure stress by asking patients to fill out questionnaires. People report their own symptoms. They tick boxes about a racing heart, sweaty palms, or sleepless nights. However, self-reporting has significant limits. Patients might hide their true feelings out of embarrassment. They might not even realise how stressed they actually are. This is why researchers are exploring a machine learning anxiety assessment. They want an objective, measurable way to evaluate stress. They need tools that look at spontaneous behaviour, rather than just physical symptoms or self-reported feelings. Standard tests can miss subtle behavioural clues. Art, on the other hand, offers a different window into the human mind. The choices we make when drawing can reveal hidden emotional states.

The Solution: Analysing art with algorithms

To test this idea, researchers collected 1044 digitised images of coloured mandalas. Mandalas are complex, circular designs often used in art therapy to promote relaxation. The team did not just look at the pictures with the human eye. They built a hybrid computer model to examine them mathematically. They combined a convolutional neural network with a Random Forest classifier. A convolutional neural network acts like a digital eye, scanning images for specific shapes and patterns. The Random Forest is a type of artificial intelligence designed to categorise data by creating multiple decision trees. The goal was to see if the computer could sort the mandala images based on the anxiety levels of the people who coloured them.

The Mechanism: Decoding colour and symmetry

The computer programme looked at four specific visual features. First, it checked colouring accuracy. Did the person stay strictly inside the lines, or did they scribble outside them? Second, it looked at colour transitions. How did they move from one shade to another across the page? Third, it measured colour entropy. This refers to the variety and randomness of the colours chosen. Finally, it checked for symmetry in the overall design. The artificial intelligence processed these patterns with impressive speed. Under strict testing conditions, the Random Forest model achieved an overall accuracy of 84.56 percent. It successfully grouped many of the images into correct categories. However, the system struggled with the high-anxiety group. It had a low sensitivity score of 0.216 for these specific cases. This happened because there were far fewer highly anxious people in the sample. The model simply did not have enough examples to learn from, creating an imbalanced dataset.

The Impact: A new support tool

This study measured visual features in digitised drawings. It suggests that specific colouring habits could act as exploratory markers for stress. The artificial intelligence cannot replace a human doctor. It is not a standalone diagnostic tool. Instead, it provides preliminary evidence that computer modelling can work alongside art-based therapy. Understanding this helps students and scientists see the future of mental health care. In the future, clinics could use this technology to flag potential issues early. A patient might simply colour a picture on a tablet in the waiting room. The software could then give the doctor an extra piece of objective information before the appointment even begins. This approach blends human creativity with computational power. It offers a highly efficient layer of defence in mental health screening.