电离辐射对细胞的影响是一个复杂的过程，取决于多个参数。癌症治疗通常涉及放射疗法的使用。除了有效杀死癌细胞外，放疗的另一个关键方面是保护健康细胞。重离子辐射由于其较高的相对生物有效性而在放射治疗领域占据着重要的地位，使其成为一种有效的治疗方法。重离子辐射的高生物效率也会对健康细胞构成危险。本研究采用统计学习方法研究了重离子辐射引起的细胞死亡程度。目的是根据可用电离辐射的物理参数来预测健康细胞的存活率。本文基于利用 PIDE 数据库进行的二次研究。在整个研究中，生成了局部回归和随机森林模型。他们的预测使用各种指标与电离辐射领域常用的线性二次模型的结果进行了比较。使用线性二次（LQM）模型和局部回归（LocReg）检查剂量和细胞存活率之间的关系。 LQM 的 R2 值为 88.43%，LocReg 的 R2 值为 89.86%。结合线性能量转移后，随机森林模型的 R2 值为 96.85%。就 RMSE 而言，线性二次模型产生 9.5910-2，局部回归产生 9.2110-2，随机森林产生 1.96 × 10-2（值越低表示性能越好）。所有这些方法也应用于对数转换的数据集，以减少数据点分布的右偏度。这显着减少了使用 LQM 和 LocReg 进行的估计（在 R2 的情况下减少了 28%），而随机森林保留了与未转换数据几乎相同的估计水平。总之，可以推断，单独的剂量对细胞存活率提供了某种令人满意的解释力，但包含线性能量转移可以在方差和解释力方面显着提高预测准确性。

The effect of ionizing radiation on cells is a complex process dependent on several parameters. Cancer treatment commonly involves the use of radiotherapy. In addition to the effective killing of cancer cells, another key aspect of radiotherapy is the protection of healthy cells. An interesting position is occupied by heavy ion radiation in the field of radiotherapy due to its high relative biological effectiveness, making it an effective method of treatment. The high biological efficiency of heavy ion radiation can also pose a danger to healthy cells. The extent of cell death induced by heavy ion radiation in cells was investigated using statistical learning methods in this study. The objective was to predict the healthy cell survival rate based on the physical parameters of the available ionizing radiation. This paper is based on secondary research utilizing the PIDE database. Throughout this study, a local regression and a random forest model were generated. Their predictions were compared to the results of a linear-quadratic model commonly utilized in the field of ionizing radiation using various metrics. The relationship between dose and cell survival rate was examined using the linear-quadratic (LQM) model and local regression (LocReg). An R2 value of 88.43% was achieved for LQM and 89.86% for LocReg. Upon incorporating linear energy transfer, the random forest model attained an R2 value of 96.85%. In terms of RMSE, the linear-quadratic model yielded 9.5910-2, the local regression 9.2110-2, and the random forest 1.96 × 10-2 (lower values indicate better performance). All of these methods were also applied to a log-transformed dataset to decrease the right skewedness of the distribution of the datapoints. This significantly reduced the estimates made with LQM and LocReg (28% decrease in the case of R2), while the random forest retained nearly the same level of estimation as the untransformed data. In conclusion, it can be inferred that dose alone provides a somewhat satisfactory explanatory power for cell survival rate, but the inclusion of linear energy transfer can significantly enhance prediction accuracy in terms of variance and explanatory power.