Multi-Output Probabilistic Prediction of Drug Side Effects Using Classical Machine Learning Algorithms.

Area of Science:

• Computational pharmacology
• Machine learning in healthcare
• Patient safety research

Background:

• Drug side effects pose significant risks to patient safety and public health.
• Traditional methods struggle to identify complex patterns linking clinical and pharmacological variables to side effects.

Purpose of the Study:

• To evaluate machine learning models for predicting multiple drug side effects probabilistically.
• To assess the utility of computational approaches in pharmacovigilance.

Main Methods:

• A cross-sectional computational study using data from 1000 medications.
• Trained and optimized Random Forest, Decision Tree, Support Vector Classifier, and KNN models.
• Utilized Chi-square tests and Principal Component Analysis (PCA) for data exploration.

Main Results:

• Significant associations identified between drug side effects and clinical condition (p < 0.05) and administered drug (p < 0.05).
• PCA revealed substantial overlap between categories, supporting a probabilistic prediction strategy.
• Tree-based machine learning models demonstrated superior performance with an accuracy of approximately 0.35.

Conclusions:

• Drug side effect prediction is a complex, multifactorial, and non-deterministic challenge.
• Probabilistic machine learning models offer a method to estimate potential adverse events.
• These models can enhance clinical decision-making and support pharmacovigilance efforts.