聯邦學習：肺癌生存率預測

Abstract

在數據保護愈發嚴格的情勢下，保險公司在遵守數據安全保護的前提下，如何利用更多的數據對於癌症險的出險、費率進行進一步預測。本文探討了一種解決企業之間數據不能相互傳輸的方式：聯邦學習。本文透過預測肺癌的存活率，比較了聯邦學習和傳統機器學習的評估效果。結果發現，聯邦學習在數據不能出本地的情況下，依舊可以達到和傳統機器學習類似的效果。因此，本文認為，聯邦學習可以在保險公司的費率、出險率的預測上提供一種新的思路，幫助保險公司克服所面臨的數據量不足，受到法規限制等問題。

Under the situation of increasingly strict data protection, it’s important for insurance companies to further predict the risk and rate of cancer insurance with more data. This paper discusses a way to solve the problem that data cannot be transmitted between enterprises—Federated learning. By predicting the survival rate of lung cancer, this paper compares the effects of federal learning and traditional machine learning. The results show that federated learning can achieve the same effect as traditional machine learning when the data must stay in local. Therefore, this paper shows that under the restriction of laws and regulations federal learning can provide a new direction in the prediction of survival rate for insurance companies to overcome the problems of insufficient data.