求職詐欺預測：應用集成學習

Abstract

求職詐欺包含透過設立假職缺蒐集個人資訊、銀行帳戶密碼或是取得金錢，本文主要探討假職缺的特徵與集成學習是否可以顯著提升預測表現，應用六種機器學習分類模型，以及職缺文字敘述、數值與虛擬變數等觀察特徵，預測樣本資料為真實或是虛假職缺。實證結果顯示，結合邏輯迴歸、K近鄰演算法、隨機森林三個子模型的集成模型表現最佳；另外，本文計算特徵重要性篩選出期望薪資平均、職缺和公司相關資訊敘述長度、職缺公告中是否含有公司商標等皆為預測求職詐欺的關鍵指標。

Job scams are fraudulent job advertisements that aim to steal personal information, banking details, or money from unsuspecting job seekers. In this article, we will be discussing the key characteristics of fake job postings and examining whether ensemble learning methods can significantly improve the performance of machine learning models in identifying job scams. We applied six different machine learning algorithms to predict fraudulent job postings using both textual and numerical variables. Our results show that the ensemble learning model, which combined logistic regression, KNeighbors classifier, and random forest classifier, performed the best. Furthermore, we used a framework based on Gini impurity to identify the ten most important factors in the random forest classifier, including average salary, company profile length, and whether the job posting had a company logo.