Artificial intelligence adaptation by the stochastic multiple regression model to determine the fibre quality of alpaca (Lama pacos)

Abstract

The application of artificial intelligence based on the multiple linear regression model with stochastic descending gradient is described in order to determine the quality of the white Huacaya alpaca fibre. In total, 1200 fibres corresponding to six alpaca samples were analysed. The fibres were characterized by optical microscopy and with the optical fibre diameter analyser (OFDA 100) equipment. Fibre diameter, medulla diameter, percentage of medullation by volume, comfort factor, and objectionable fibres were considered as independent variables, and the “Soft” factor was considered as a response variable. This last variable resulting from the difference in the comfort factor and objectionable fibres served to give a logical order to the data matrix and obtain an accurate prediction model. The average values were 26.80 ± 6.95 for the fibre diameter, 14.10 ± 5.92 for the medulla diameter, 24.75 ± 13.20 µm for the percentage of medullation by volume and 71.56 ± 13.04% for the comfort factor. The machine learning multiple linear regression modelling fitted a small sample size with high precision, showing minimal errors, and optimized with the stochastic gradient descent algorithm predicted a Soft factor very close to the observed Soft factor. It is concluded that the multiple linear regression technique with the stochastic approach satisfies the prediction of the new factor called "soft" and that it represents the appropriate modelling for the prediction of fibre quality in the textile industry.