Comparison of Effectiveness of Machine Learning Methods in Predicting Chemical Compound Toxicity Enhance Pharmaceutical Product Safety

Abstract

This study compares the effectiveness of machine learning methods in predicting the toxicity of chemical compounds using a dataset containing 5,000 samples with 14 key features. The dataset underwent preprocessing, including normalization, missing data handling, and oversampling to address data imbalance. The models used include Decision Tree, Random Forest, Extra Trees, and Gradient Boosting, validated using k-fold cross-validation. Evaluation based on accuracy, precision, recall, and F1-score showed that Gradient Boosting achieved the best performance with 92.3% accuracy, though it still faces challenges such as overfitting and interpretability limitations. Compared to in vitro and in vivo methods, machine learning is more efficient but still requires further experimental validation. This study recommends optimizing models through ensemble learning and explainable AI to improve prediction reliability.