A hybrid approach for pattern recognition and interpretation in age-related false memory

Area of Science:

• Cognitive Neuroscience
• Computational Psychology
• Gerontology

Background:

• Aging impairs cognitive functions like memory, impacting quality of life.
• Mechanisms of age-related false memory formation are poorly understood.
• Current methodologies have limitations in studying cognitive aging.

Purpose of the Study:

• Develop a hybrid computational method to distinguish true from false memories in younger and older adults.
• Identify behavioral patterns associated with age-related memory decline.
• Enhance the interpretability of findings using language models.

Main Methods:

• A hybrid approach combining machine learning (LightGBM) and language models.
• Identification of nine key features using permutation importance.
• Examination of feature interactions with age for cognitive decline insights.

Main Results:

• Younger adults used reading time to reject misleading information (lures).
• Older adults showed increased vulnerability to inference from semantic similarity.
• The modified LightGBM model achieved an F1-score of 0.82 and recall of 0.88.

Conclusions:

• Hybrid computational methods are effective in uncovering memory decline patterns.
• Findings provide insights into false memory mechanisms in aging.
• The developed model outperforms deep learning and transformer-based approaches.