Colour is used to describe the quality of a product during quality control and assurance as it associates with the chemical, biochemical, microbial, and physical properties changes that happen through the processing and degradation by lights or chemical exposure. However, the colour of a product is normally measured by expensive instruments like spectrophotometer, and colorimeter, and they regularly need maintenance, services, and repairs. Furthermore, they only provide the data for the calculation of colour index including the yellowness index (YI), and not the YI value. To address the abovementioned limitations, a locally weighted Kernel partial least square regression (LW-KPLSR) which is a soft sensor that is cheap and can perform real-time measurement could be used to predict the YI directly. Therefore, this study aims to examine the performance of LW-KPLSR to predict the YI. Hence, standard yellow colour charts listed with the red, green, and blue values, as well as other data were used as the data for model development and validation. Meanwhile, the results from LW-KPLSR were compared with principal component regression, partial least square regression, and locally weighted partial least square regression models. It was found that the LW-KPLSR gave 36% to 695% lower root means square errors and mean absolute percentage error values for both training and testing data as compared to the other models. Meanwhile, the coefficient of determination for LW-KPLSR, especially for the testing data was also much higher than the rest of the models. Additionally, its error of approximation was 48% to 638% lower than other models. [ABSTRACT FROM AUTHOR]