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

Introduction: Aging is associated with a decline in essential cognitive functions such as language processing, memory, and attention, which significantly impacts the quality of life in later years. Despite the serious consequences of age-related cognitive decline, particularly in the formation of false memories, the underlying mechanisms remain poorly understood. This knowledge gap is partly due to limitations in current methodologies used to examine age-related cognitive changes and their origins. Methods: In the present study, a hybrid approach was developed that combines optimized machine learning techniques with large-scale transformer-based language models to identify behavioral patterns distinguishing true from false memories in both younger and older adults. The best-performing model, a modified version of the Light Gradient Boosting Machine (LightGBM), identified nine key features using permutation importance. Feature interactions with age were further examined to understand their relationship with cognitive decline. Additionally, the modified LightGBM was integrated with a language model to enhance interpretability. Results: The findings revealed that younger adults benefited from target encoding time during reading, which helped them correctly reject misleading information (lures), whereas older adults were more vulnerable to inference caused by semantic similarity. Discussion: These results offer important insights into the mechanisms of false memory in aging populations and demonstrate the utility of hybrid computational methods in uncovering behavioral patterns related to memory decline. The modified LightGBM achieved the highest overall performance with an F1-score of 0.82 and recall of 0.88, outperforming all evaluated deep learning and transformer-based models.