Drug Toxicity Classification Based on ReliefF and K-means Algorithm

Mongolian medicine Garidi-5 is a well-known regional compound formulation used for drying and treating the disease of yellow water. It contains the toxic herb Cao Wu and has been in use till date. Biochemical indicators in mice are important for studying drug toxicity. In this study, 32 SD rats were divided into four groups and given a daily oral dose of Garidi-5 at 0 g/kg, 0.3429 g/kg, 0.0857 g/kg, and 0.0214 $g$ /k $g$ , respectively, for 28 consecutive days. A feature selection algorithm based on the ReliefF algorithm was used to identify Garidi-5 poisoning in rats using organ indicators and blood biochemical indicators. In terms of feature selection, the most important relevant indicators were CKMB, ALP, HWR, CREP, CK, UREA, ALT, and L2WR. The K-means algorithm was used to perform cluster analysis on the rats' data indicators, and the results showed that after using the ReliefF algorithm for feature selection, the predicted accuracy and silhouette coefficient increased by 21.43% and 0.0779, respectively, when the feature weight threshold was set to 0.035. Empirical analysis showed that the ReliefF feature selection algorithm can improve the accuracy of toxicity prediction.