基于机器学习的老年人诈骗预警

Machine Learning Based Fraud Warning for the Elderly

摘要: 全球范围内，我们正经历着显著的人口老龄化现象，这主要表现为人们的预期寿命持续增长。这种现象并非偶然，而是人口发展的一种自然演进。在中国，这一趋势尤为明显，这是因为我们的老年人口比例正在迅速上升。中国正面临着显著的人口老龄化现象，其规模之巨以及持续增长的速度令人瞩目。预计在未来的几十年里，这个趋势将持续下去。随着社会经济的进步和多元化的影响，一些问题逐渐浮出水面，老年人诈骗已经成为非常严重的现象。同时，老年人也因为自身的多种因素，例如：防范意识弱、容易轻信他人、缺乏社会支持和科学知识，容易成为诈骗者的行骗对象。因此，本文选取CHARLS数据库中2020年的部分针对多因素对老年人被诈骗进行分类分析，共计12个变量：是否被诈骗过、性别、学历、年龄、个人经济情况、满意度、健康状况、认知能力、抑郁程度、记忆能力、孤独感、保险保障。数据预处理部分使用excel完成，选用XGBoost、GBDT、随机森林共三种模型对老年人被诈骗的数据进行建模研究，并进行比较，选出最优模型。其中，GBDT模型的准确率为0.93，F1值为0.93，为最优模型。进而，本文分析不同因素对老年人被诈骗的影响机制，建立预警模型，提出有针对性的政策措施，提高老年人的防范意识，降低对财产、心理等的损害。

Abstract: On a global scale, we are experiencing a significant aging population, which is mainly reflected in the continuous increase in people’s life expectancy. This phenomenon is not accidental, but a natural evolution of population development. In China, this trend is particularly evident because the proportion of our elderly population is rapidly increasing. China is facing a significant aging population, with its enormous scale and sustained growth rate being remarkable. It is expected that this trend will continue in the coming decades. With the progress of socio economy and the impact of diversification, some problems have gradually surfaced, fraud among the elderly has become a very serious phenomenon. At the same time, elderly people are also prone to becoming targets of scammers due to various factors, such as weak awareness of prevention, easy trust in others, lack of social support and scientific knowledge. Therefore, this article selects parts of the CHARLS database from 2020 to classify and analyze elderly people who have been defrauded based on multiple factors, with a total of 12 variables: whether they have been defrauded, gender, education level, age, personal economic situation, satisfaction, health status, cognitive ability, depression level, memory ability, loneliness, and insurance coverage. The data preprocessing part was completed using Excel. XGBoost, GBDT and random forest models were used to model and study the data of elderly people being defrauded, and the optimal model was selected through comparison. Among them, the accuracy of the GBDT model is 0.93 and the F1 values are 0.93, making it the optimal model. Furthermore, this article analyzes the impact mechanisms of different factors on elderly people being defrauded, establishes an early warning model, proposes targeted policy measures, enhances elderly people’s awareness of prevention, and reduces damage to property, psychology, and other aspects.