This article mainly uses sparse Global Positioning System trajectory data to identify traffic travel pattern. In this article, the data are preprocessed and the eigenvalues are calculated. Then, the Global Positioning System track points are identified and extracted by walking and non-walking segments. Finally, the three machine learning models of support-vector machine, decision tree, and convolutional neural network are used for comparison experiments. The innovation of this article is to propose a walking and non-walking identification method based on density-based spatial clustering of applications with noise clustering. The method takes into account the continuous state between the geographical distributions, and it has better noise immunity against the influence of external factors. In this process, this article directly achieves better conversion point recognition results through the Global Positioning System track point information, which lays a good foundation for the accuracy of travel pattern recognition. The experimental results of this article show that compared with threshold-based and multi-layer perceptron–based methods, the recognition method based on density-based spatial clustering of applications with noise clustering has the highest accuracy, reaching 82.20%. Then, support-vector machine, decision tree, and convolutional neural network are used for traffic travel pattern recognition. The F1-score of support-vector machine is the highest, reaching 0.84, and the F1-scores of decision tree and convolutional neural network are 0.78 and 0.80, respectively. Finally, the support-vector machine was used for the recognition test to achieve an accuracy of 76.83%.

中文翻译：

---

基于稀疏全球定位系统轨迹数据的交通出行模式识别

本文主要使用稀疏的全球定位系统轨迹数据来识别交通出行模式。在本文中，对数据进行了预处理并计算了特征值。然后，通过步行和非步行段识别和提取全球定位系统轨迹点。最后，使用支持向量机、决策树和卷积神经网络这三种机器学习模型进行对比实验。本文的创新之处在于提出了一种基于密度空间聚类的步行和非步行识别方法，应用噪声聚类。该方法考虑了地理分布之间的连续状态，对外界因素的影响具有较好的抗噪能力。在这个过程中，本文通过全球定位系统轨迹点信息直接实现了更好的转换点识别结果，为出行模式识别的准确性奠定了良好的基础。本文的实验结果表明，与基于阈值和基于多层感知器的方法相比，基于密度的空间聚类与噪声聚类应用的识别方法具有最高的准确率，达到82.20%。然后，使用支持向量机、决策树和卷积神经网络进行交通出行模式识别。支持向量机的F1-score最高，达到0.84，决策树和卷积神经网络的F1-score分别为0.78和0.80。最后，