The process of assigning grammatical categories, such as ``Noun'' and ``Verb,'' to every word in a text corpus is known as part-of-speech (POS) tagging. This technique is widely used in applications like sentiment analysis, machine translation, and other linguistic and computational tasks. However, the unique features of the Pashto language and its limited resources present significant challenges for POS tagging. This study explores the critical role of POS tagging in the Pashto language by employing six popular deep-learning and machine-learning techniques. Experimental results demonstrate machine learning methods' effectiveness in capturing Pashto text's grammatical patterns. The evaluation is based on a well-curated and annotated dataset of Pashto text, meticulously compiled from diverse sources and enriched with POS tags, providing a reliable foundation for performance analysis. Among the tested algorithms, K-Nearest Neighbor (KNN) and Decision Tree achieved the highest accuracy rates, with 94.19% and 94.34%, respectively. Random Forest and Support Vector Machine (SVM) also delivered competitive results, exceeding the 90% accuracy threshold. Multi-Layer Perceptron (MLP), evaluated with various activation functions like ReLU and Tanh, achieved an accuracy of 87.25%, while Naïve Bayes, tested with different variants such as Multinomial NB and Gaussian NB, attained 83.33%. These results highlight the potential of machine learning techniques in overcoming the challenges associated with Pashto POS tagging.