Predictive Modelling of Clinical Trial Toxicity by Using Cuckoo Search-Ensemble Method

Toxicity prediction in drug development is pivotal to determine the clinical suitability of new compounds. This study aims to enhance the accuracy of toxicity prediction by integrating an ensemble method, specifically Random Forest and AdaBoost, with the Cuckoo Search algorithm for optimized feature selection. The primary research objective is to develop a robust and efficient predictive model that improves the effectiveness of clinical toxicity classification by leveraging advanced feature selection and ensemble learning techniques. This novel approach was tested using the Clintox dataset, comprising 1431 chemical compounds, aiming to predict clinical toxicity. The methodology involved feature selection to eliminate irrelevant attributes, followed by hyperparameter tuning to optimize model performance. The use of the Cuckoo Search-Ensemble model demonstrated a significant improvement in predictive capability, showcasing robustness against data variations. The model's effectiveness is further underscored by feature selection and hyperparameter tuning, which are crucial for enhancing performance. The evaluation of the model utilized critical performance metrics including accuracy, precision, recall, and F1-score. Our results found good performance for the AdaBoost model using a learning rate of 0.1 and 150 estimators, where it achieved an F1-Score of 0.6515 and an accuracy of 0.9125.