Machine Learning in Psychology: part 1

Building on last week’s blog post about mental health, the future applications of data science in psychology will be tremendous. As I delve deeper into research on machine learning and artificial intelligence, I discover more about how scientific advancements are being made in psychology. Today, however, I will discuss one of the articles included in my research paper about the role of machine learning in academic research in cognitive psychology. By examining this article in depth, I hope readers will gain a better understanding of the applications of machine learning in psychology.

In a 2022 study, Valentina Bachurina and her team explored how machine learning could be used to predict the cognitive behavior of their test subjects. The primary goal of the experiment was to examine the correlation between eye movements and attention, as where we look often aligns with what we are thinking about. Therefore, our eye movements play a crucial role when we think and concentrate.

In the experiment, 62 adult participants were given two different color-matching tasks with six levels of difficulty. One task involved colored balloons, and the other involved colored clowns. In the balloon task, participants observed a color presented on the screen and pressed a button if the color matched the previous one. The task became more challenging as participants were given more colors to remember. Their performances were measured by response speed and accuracy. Eye-tracking data from the experiment was then used for machine learning analysis, which was the crux of the study.

Several machine learning models were used to predict cognitive performance, including Linear Regression, Lasso Regression, Ridge Regression, K-Nearest Neighbors (KNN) Regression, XGBoost Regression, and Random Forest Regression. Each model had its own significance, which I will explore in more depth in the next post. In short, Linear Regression provided a simple, interpretable baseline; Lasso and Ridge Regression offered regularization to prevent overfitting and highlight key features; KNN captured non-linear relationships; and XGBoost and Random Forest, as ensemble methods, delivered high predictive accuracy and identified important features influencing performance. Together, these models allowed the researchers to determine how task difficulty, reaction time, and eye movements could predict cognitive performance.

In summary, task difficulty and reaction time were strong predictors of behavior, whereas eye movements alone were weaker predictors. However, specific eye-tracking indices effectively predicted outcomes. Key features included the number of fixations (how often the gaze settled), the number of saccades (quick eye movements), fixation duration (how long the gaze stayed on one point), and pupil size (varying with cognitive load). These elements offered valuable insights into participants’ cognitive processes and mental attention during the tasks.

Ultimately, this experiment demonstrates that the use of machine learning in psychology is becoming increasingly prevalent due to its ability to analyze large datasets and identify unrecognized patterns to make predictions. As machine learning continues to develop, the field of psychology continues to shift from the idea of explaning behavior to predicting behavior.