NIRS Calibration Models for the Prediction of Textural Attributes of Eba

The experiment conducted in IITA, Nigeria, on NIRS model development for the prediction of textural attributes of Eba is reported in this document. Texture has been identified as a key quality trait that influences the consumer acceptability of Eba. Therefore, developing a high throughput technique that can be employed to evaluate texture rapidly will significantly benefit the breeders. The NIRS prediction model is a relatively cheap and fast technique that can screen samples quickly. In this experiment, NIRS models were developed to predict textural attributes of Eba samples which were prepared from forty Gari samples using standardized protocols. WP4 supplied the cassava genotypes used for producing Gari samples using the RTBfoods SOP (doi.org ). The instrumental texture of the Eba samples was carried out using Stable Micro Systems TA.XTplus texture analyzer while the NIRS spectra data was also collected using a NIRS spectrophotometer (FOSS XDS Rapid content analyzer). Pre-treated of the spectra data was done using Standard Normal Variate Detrend (SNVD) and Savitzky Golay mathematical treatments. At the same time, the model was developed using a modified partial least square (MPLS) algorithm. In the developed model, cohesiveness had the highest R2cal of 0.94, followed by hardness with R2cal of 0.79, but the lowest R2cal was observed for gumminess (0.21). The models were validated using independent samples, and the coefficient of determination in prediction (R2pred) for chewiness and hardness were 0.58 and 0.54, respectively. The model developed in this experiment can be used for the routine prediction of the hardness of Eba. The models will also be further improved by introducing more data sets into the calibration and validation data.