基于特征性脂肪酸和甘油三酯指标的油茶籽油掺伪定性鉴别模型对比分析

为解决油茶籽油掺伪其他植物油的定性鉴别问题，在油茶籽油中分别掺入大豆油、花生油、葵花籽油、棉籽油、葡萄籽油、菜籽油、棕榈油和米糠油，设置高和低两种不同掺伪梯度，基于14个特征性脂肪酸和甘油三酯指标，运用Python语言构建并对比分析了二分类决策树模型、多分类决策树模型和多层感知机人工神经网络(MLP-ANN)模型用于油茶籽油掺伪定性鉴别的效果。结果表明：高和低掺伪梯度下，二分类决策树模型对油茶籽油掺伪其他植物油的定性鉴别的准确率均达到0.95以上；多分类决策树模型的精确率和准确率在高掺伪梯度下均达到了0.95,但在低掺伪梯度下仅为0.90;在高和低掺伪梯度下，MLP-ANN模型对油茶籽油掺伪定性鉴别的平均精确率均达到0.98,准确率分别达到0.97和0.98。相比于决策树模型，MLP-ANN模型能很好地实现油茶籽油掺伪定性鉴别。

Comparative analysis of qualitative identification models for oil-tea camellia seed oil adulteration based on characteristic fatty acid and triglyceride

In order to solve the qualitative identification problem of adulterated oil-tea camellia seed oil with other vegetable oils, soybean oil, peanut oil, sunflower seed oil, cottonseed oil, grape seed oil, rapeseed oil, palm oil and rice bran oil were mixed into oil-tea camellia seed oil respectively, two different adulteration gradients of high and low were set up, and based on characteristic fatty acid and triglyceride indicators, the effects of the binary decision tree model, multi-classification decision tree model and multilayer perceptron artificial neural network (MLP-ANN) model for qualitative identification of adulterated oil-tea camellia seed oil were compared and analysed using Python language. The results showed that the accuracy of the binary decision tree model for qualitative identification of oil-tea camellia seed oil adulterated with other vegetable oils under high and low adulteration gradients was above 0.95. The accuracy and precision of the multi-classification decision tree model reached 0.95 at high adulteration gradient, but only 0.90 at low adulteration gradient. Under high and low adulteration gradients, the average precision of MLP-ANN model for qualitative identification of adulterated oil-tea camellia seed oil reached 0.98, and the accuracy reached 0.97 and 0.98 respectively. Compared with the decision tree model, the MLP-ANN model can well realize the qualitative identification of adulterated oil-tea camellia seed oil.