Using correlation matrices to standardise sweet liking status classification

•Misclassification occurs when AHC alone is used to determine Sweet Liking Status.•Correlation matrix can remove noise in Sweet Liking Status classification.•Correlation matrix with validated liking scores is recommended for future studies.•SLS can be determined by the rapid classification method using 3 sucrose solutions. Distinct hedonic patterns of sweet taste liking have been widely recognised for more than half a century. Despite there being a growing consensus on the role of Sweet Liking Status (SLS) in food choice behaviour, current classification methods for this phenotype generally lack consistency. Using a large dataset (n = 865), the present study applied Agglomerative Hierarchical Clustering (AHC) followed by correlation matrices as a validated and robust method for SLS classification by using five sucrose solutions (3, 6, 12, 24 and 36 %). As demonstrated in the present study, AHC alone was not a sufficient method to generate reliable SLS clusters. Following a validated correlation matrix approach, three distinct consumer clusters were identified: High Sweet Likers (HSL), Medium Sweet Likers (MSL) and Low Sweet Likers (LSL). Robust mean liking scores were generated for each of the three clusters across five different concentrations of sucrose. The results suggested that in order to enable more efficient and comprehensive SLS classification, a correlation-based approach for SLS classification using the validated liking means provided in the current study should be adopted in future research. In addition, a rapid three-solution method (3 %, 12 % and 36 %) was also explored as a simplified and more efficient way of classifying participants for SLS. The rapid three-solution method accurately classified the majority of HSL, MSL and LSL within the dataset. The data showed a good level of agreement between the rapid three-solution method and validated five-solution method, therefore suggesting that a rapid three-solution method can be considered when exploring the two hedonic extremes (HSL and LSL) when additional noise in the data can be tolerated.