Drug-drug interactions (DDIs) are a major healthcare concern because combining multiple medications canlead to adverse reactions, reduced treatment effectiveness, and increased clinical risks. This study presentsa machine learning approach for classifying drug-drug interaction types using pharmacological text dataobtained from the DrugBank database. The proposed approach applies machine learning models including Logistic Regression, Linear SupportVector Machine (SVM), and Random Forest for automated multi-class interaction type classification basedon pharmacological interaction descriptions. TF-IDF vectorization, a Natural Language Processing (NLP)technique, was used to transform textual interaction descriptions into numerical feature representationssuitable for machine learning classification. The transformed data were then used to train and evaluate theclassifiers. Model performance was primarily evaluated using the Macro F1-score, while Accuracy,Weighted F1-score, confusion matrix analysis, and learning curve analysis were used to support the results. The results showed that Linear SVM achieved the strongest overall classification performance with a MacroF1-score of 0.855 and an Accuracy of 0.879. Logistic Regression achieved a Macro F1-score of 0.815 andan Accuracy of 0.863, while also providing improved interpretability and probability-based outputs. Incontrast, Random Forest achieved a Macro F1-score of 0.689 and an Accuracy of 0.735, demonstratingcomparatively lower performance and reduced generalization capability. The findings demonstrate that combining TF-IDF vectorization with machine learning classificationmethods provides a simple, interpretable, and computationally efficient approach for drug-drug interactiontype classification using pharmacological text data. A limitation of the study is its reliance on textualinteraction descriptions from a single data source without incorporating additional pharmacologicalfeatures. The study contributes to pharmacological text classification research by demonstrating that strongclassification performance can be achieved while maintaining interpretability and generalization capability.