Estimation of socioeconomic attributes from location information
Estimation of socioeconomic attributes from location information
复制标题
DOI：
10.1007/s42001-020-00073-w
复制
发表时间：
2020-04
影响因子：
3.2
通讯作者：
Shohei Doi;T. Mizuno;N. Fujiwara
中科院分区：
文献类型：
--
作者：
Shohei Doi;T. Mizuno;N. Fujiwara
及时估计社会经济属性的分布及其变动对于学术、行政和营销目的至关重要。在本研究中，假设个人属性影响人类行为和运动，我们根据位置信息预测这些属性。首先，我们通过监督学习方法预测个体的社会经济特征，即逻辑套索回归、高斯朴素贝叶斯、随机森林、XGBoost、LightGBM和支持向量机，使用我们收集的个人属性和访问特定设施频率的调查数据来检验我们的猜想。我们发现，正如现有研究所做的那样，性别是一个关键属性，从地点到社交网络服务等其他来源都是高度可预测的。其次，我们将使用调查数据训练的模型应用于实际的 GPS 日志数据，以检查我们的方法在现实环境中的性能。尽管我们的方法对于调查数据的表现不佳，但结果表明我们可以从 GPS 日志推断性别。
Timely estimation of the distribution of socioeconomic attributes and their movement is crucial for academic as well as administrative and marketing purposes. In this study, assuming personal attributes affect human behavior and movement, we predict these attributes from location information. First, we predict the socioeconomic characteristics of individuals by supervised learning methods, i.e., logistic Lasso regression, Gaussian Naive Bayes, random forest, XGBoost, LightGBM, and support vector machine, using survey data we collected of personal attributes and frequency of visits to specific facilities, to test our conjecture. We find that gender, a crucial attribute, is as highly predictable from locations as from other sources such as social networking services, as done by existing studies. Second, we apply the model trained with the survey data to actual GPS log data to check the performance of our approach in a real-world setting. Though our approach does not perform as well as for the survey data, the results suggest that we can infer gender from a GPS log.