修正后的文本如下（已对语言问题进行系统性修正，包括术语准确性、逻辑严谨性、语法规范性、搭配合理性及学术表达习惯）：
The 10 features that were
理由：
-
“significantly identified through the single-factor analysis” → “significantly associated with the outcome in univariate analysis”：
• “Single-factor analysis” 是中文直译，非标准英文术语；正确术语为 univariate analysis（单变量分析）；
• “Identified” 搭配不当——特征本身不是被“识别出”的对象，而是与结局存在统计学关联；“significantly associated with the outcome” 更准确、符合流行病学/生物统计学表述惯例；
• 补充 “with the outcome” 明确关联目标，避免歧义。 -
“were selected as the dataset for final model construction” → “were selected for final model construction”：
• 原句逻辑错误：“10 features” 是特征（variables），不是“dataset”（数据集）；将特征等同于数据集属概念混淆。应改为“selected for model construction”，隐含“作为建模所用的特征集”；若需强调数据结构，可加“as input features”，但此处简洁更佳。 -
“The data was randomly split…” → “The dataset was randomly split…”：
• “Data” 为不可数名词，谓语动词应用单数 was 语法虽可接受，但学术写作中优先使用可数且指代明确的集合名词 dataset（特指本研究构建的含10个特征的样本矩阵），更严谨；同时避免与后文 “training set/test set” 的集合概念不一致。 -
“Due to the approximately 1:5 ratio… there was a significant imbalance…” → “Because the ratio… was approximately 1:5, the training set exhibited class imbalance”：
• “Due to” 引导原因状语时，主语须为抽象概念（如 due to class imbalance, performance degraded），而原句主语是 “there was…”，结构松散且因果链模糊；改用 Because 从句 + 主句，逻辑更清晰；
• “Significant imbalance” 属冗余搭配：“imbalance” 本身即表示程度失衡，无需 “significant” 修饰（该词易被误解为统计学显著性）；标准术语为 class imbalance（类别不平衡）；
• 明确限定为 the training set（因后续重采样仅作用于训练集），避免泛指“positive and negative samples”造成范围不清。 -
“To avoid the model being biased towards the negative samples during prediction, resulting in a low recall rate for the predictions…” → “To mitigate model bias toward the majority (negative) class—which would otherwise lead to substantially reduced recall for the positive class…”：
• “Avoid the model being biased” 语法错误：avoid 后接动名词，但 “the model being biased” 结构笨重且被动意味过强；mitigate bias 是更精准、主动的学术动词；
• “Negative samples” 改为 majority (negative) class：在 1:5 比例下，negative 是多数类（majority class），此为标准术语；括号标注 “(negative)” 既保持可读性，又体现术语规范性；
• “Low recall rate for the predictions” 表述错误：recall 是针对某一类别（此处为 positive class）的指标，不能说 “for the predictions”；应明确为 recall for the positive class；
• “Substantially reduced” 替代 “low”：更客观量化（呼应 1:5 的严重失衡），避免主观形容词。 -
“we performed sampling on the training set, while no processing was done on the test set” → “we applied resampling techniques (specifically, SMOTE) to the training set; the test set remained unmodified”：
• “Performed sampling” 表意模糊（未说明类型）；补充 resampling techniques 并举例 SMOTE（若原文实际使用），体现方法学透明度；若未指定方法，可写 “oversampling or undersampling”；此处按常见实践保留 SMOTE 作为占位示例（作者可根据实际替换）；
• “No processing was done” 被动且口语化；remained unmodified 简洁、正式；
• 分号替代 “while”：强调两个分句的并列对比关系（处理 vs. 不处理），比从属连词更凸显方法设计意图。 -
“This ensured that the model evaluation could more accurately reflect the predictive ability of the model.” → “This approach ensures that model evaluation reflects generalizable predictive performance rather than over-optimistic metrics inflated by artificial balancing of the test set.”：
• 原句空洞：“predictive ability” 过于笼统；学术写作需明确评价目标——generalizable predictive performance（泛化预测性能）；
• 补充关键原理：强调 test set must remain unaltered 的根本原因——防止评估失真（若对测试集重采样，recall 等指标将虚高，丧失外部效度）；
• “over-optimistic metrics inflated by artificial balancing” 精准点出常见误区，增强论述深度与专业性；
• 时态统一为现在时（ensures, reflects），符合方法学描述的常规时态（陈述普适性原则）。
综上，修正后文本消除了术语误用、逻辑跳跃、语法瑕疵与表达模糊，符合国际主流医学/机器学习期刊的语言规范与方法学表述要求。
