This project compared deep learning models and their hybrids with linguistic features in detecting AI-generated text. The results show that most deep learning models improved their performance when combined with linguistic features, forming hybrid models. The study also tested conventional machine learning models on TF-IDF-PCA features and on linguistic features extracted from the text to provide additional points of comparison. The results showed that linear models such as SVM and logistic regression tend to work well with TF-IDF-PCA data, while tree-based models excel with linguistic feature data. Overall, deep learning models and their linguistic hybrids outperformed the machine learning models. Among deep learning models, transformers (such as BERT and RoBERTa) and their linguistic hybrids outperformed conventional deep learning models (such as CNNs and LSTMs) and their linguistic hybrids. The study also examined whether AI-generated and human-written texts differ in their linguistic characteristics.