Research on Apple Origins Classification Optimization Based on Least Angle Regression in Instance Selection

Abstract

Abstract The combined near-infrared spectral analysis technology and machine learning is widely used in fruit quality detection. The train set split from the heavy sample set contains redundant samples, and modeling directly may cause larger errors. Instance selection methods can be used to improve the performance of the raw train set by optimally selecting the effective samples. So, the least angle regression-based instance selection (LARIS) method was proposed to improve the model accuracy of fruits classification in this paper, an experiment was conducted on a total of 952 apples from four origins. Before splitting, the spectral reconstruction methods were used for preprocessing and the 19 outliers were eliminated by statistics. The sample set partitioning based on joint x-y distance (SPXY) was used to split the sample set into the raw train set of 699 samples and the test set of 234 samples. 511 samples with a compression ratio of 26.90% and the random train set with the same compression ratio were built based on training samples. Compared these classifiers trained by three train sets, the model’s accuracy established by the optimal train set is 96.6%, which are 4.7% and 6.4% higher than the raw and random train sets. And the average precision and recall of four origins are higher 6% than the raw and random train set. Therefore, the prediction accuracy of apple origins classification model is improved by LARIS. The LARIS method enriches the application and it provides an experimental support for the least angle regression algorithm in instance selection.