To accurately identify vehicle lane change intentions and enhance driving safety, this paper utilizes the CitySim dataset to conduct research on lane change intention recognition. First, the driving intentions of target vehicles are categorized into three types: going straight (TS), lane changing to the left (TL), and lane changing to the right (TR). A dimensionality reduction strategy is then employed to eliminate redundant data. Next, lane change intention recognition algorithms are designed based on fundamental principles of K-Nearest Neighbors (KNN), Random Forest (RF), Convolutional Neural Networks (CNN), and Multi-Layer Perceptrons (MLP). Subsequently, five-fold cross-validation is used for hyperparameter optimization of the KNN and RF models, employing techniques such as grid search. For the CNN and MLP models, relevant functions are introduced to dynamically adjust the learning rate and increase the number of training epochs. Finally, to prevent overfitting in the field, Synthetic Minority Over-sampling Technique (SMOTE) is utilized to balance the samples. The results indicate that all four models achieve high accuracy in lane change intention recognition, with the accuracy for TS exceeding 97%. Compared to the other three models, the accuracy of the KNN model is higher by 1, 2, and 3 percentage points, making it the most accurate overall.