Combinatorial Long-Short Term Memory– Group Interaction in Contribution Method for Estimation of Lethal Dose of Pesticides
Main Article Content
Abstract
Humans are continuously exposed to a variety of chemicals , including pesticides , many of which are potentially toxic and have carcinogenic effects . Determining the human toxicity of chemicals remained a challenge due to the large resources required to evaluate a chemical in vivo. In this study , a hybrid model that combined the long-short term memory (LSTM) and Group Interaction Contribution (GIC) was used to predict the LD50 toxicity in rats with chemical pesticides , which was developed using a database relatively large consisting of 303 pesticides belonging to different chemical groups . The architecture of LSTM-MLR hybrid model was carefully selected by testing different number of hidden neurons and different number of training iterations in order to avoid the model over fitting . The parameters selection process revealed that the model should be developed using 300 hidden neurons and trained with 1000 iterations . The developed model with best selected parameters achieved a very interesting results and a high accuracy in the prediction of LD50 in testing phases with 0.8888 , 0.1312 , 0.3622 , 0.2926 , 20.64 , and 39.99, for R² , MSE , RMSE , MAE , RAE, and RRSE , respectively.