Dementia prediction remains a complex challenge. Recently, machine learning (ML) techniques have been increasingly used to improve prediction accuracy based on demographic, genetic, and neuropsychological measures. In this study, we apply a modified one-boundary diffusion model to data from the EPIC-Norfolk Prospective Population Cohort Study (2021), which includes 7,171 participants aged 48–92, 8% of whom developed dementia over a 10-year follow-up period. Participants completed the Visual Sensitivity Task (VST), a perceptual detection task designed to assess visual processing speed. A recent study by Begde et al. (2024) reported a potential relationship between visual sensitivity and dementia risk. Building on this work, we use trial-by-trial VST data to estimate cognitive model parameters increase in drift rate (reflecting visual processing speed), boundary, and non-decision time, and examine their associations with demographic variables and subsequent dementia outcomes. We conducted survival analysis in three conditions: demographics with mean RT, with RT quantiles, and with cognitive parameters. The increase in drift rate (dv) (0.76 [0.68–0.86], p < 0.0001), boundary (a) (1.11 [1.01–1.23], p = 0.035) , and the interaction between age and increase in drift (1.18 [1.07–1.30], p= 0.001) were significant. Additionally, supervised ML algorithms were trained under three conditions: demographic information, cognitive parameters, and RT quantiles to distinguish dementia from cognitively healthy individuals. Despite the class imbalance, logistic regression classifier showed a precision of 0.25, specificity of 0.84, and sensitivity of 0.52. This study has three goals: (1) to fit a cognitive model to a large, population-based dataset, (2) to assess the relationship between demographic variables, cognitive parameters, and dementia diagnoses, and (3) to evaluate how well these parameters contribute to the prediction of future dementia risk.