Since the proximate analysis of biomass is more convenient and costs lower than the ultimate analysis and chemical analysis

using the proximate analysis result to predict high calorific value of the biomass is of great significance. The process of using machine learning to fit dependents strongly on the data itself(the size and quality of the dataset)

a small number of poor quality data will have a huge influence on the fitting effect. So

it is necessary to explore a linear fitting method suitable for the case of small amount of data and the existence of outliers at the present stage. On this basis

three methods for handling the abnormal points are put forward: method I uses quartile distance to find the abnormal points and eliminate them

method II preprocesses the original data by using the statistical value of the intermediate data

method II uses ridge regression instead of the conventional linear regression to enhance its ability to resist interference of abnormal points. The results show that

the ridge regression has the best linear fitting effect under the condition of small amount of data and abnormal points at present.