MaxPrestige/Synthetic-Diabetes-Dataset

---
license: mit
task_categories:
- token-classification
- reinforcement-learning
language:
- en
tags:
- health
- diabetes
- glucose
- classification
size_categories:
- 10K<n<100K
---
# ⛽ Synthetic Diabetes Data
This dataset contains various features that are helpful for predicting if a patient has diabetes. The data is compiled into a single csv file for analysis and model training.

# 📁 Dataset Description
Contains synthetic data on based on synthetic patient information.

# Columns
The dataset includes the following columns:

- abdominal_obesity (int)

- alcohol_consumption_per_week (int):
> The amount of alcohol consumption per week of the patient.

- alcohol_group (string):
> The category the patient is placed in, in regards to their alcohol consumption.

- bmi (float):
> The body mass index of the patient.

- bmi_group (string):
> The category the patient is placed in, in regards to their bmi.

- cardiovascular_history (int):
> Indicates whether or not patient has a history of cardiovascular.

- cholesterol_total (int):
> Sum of all cholesterol types in blood. Indicator of diabetic dyslipidemia, referring to a specific pattern of abnormal blood lipid levels commonly seen in people with diabetes.

> Total Cholesterol = HDL+LDL+VLDL = HDL+LDL+(Triglycerides/5)
> Ideal Level: Total Cholesterol: <200 mg/dL

- diabetes_risk_score (float)

- diabetes_stage (string)

- diastolic_bp (int):
> Pressure when the heart relaxes between beats.

> Elevated BP is common in diabetes but does not diagnose it.

> Useful for risk stratification, not for diagnosis.

- education_level (string):
> The education level of the patient.

- employment_status (string):
> The employment status of the patient.

- ethnicity (string):
> The ethnicity of the patient.

- family_history_diabetes (int):
> Indicates whether or not the patient's family has history of diabetes.

- gender (string):
> The gender of the patient.

- glucose_fasting (int):
> Glucose levels when fasting

- glucose_postprandial (int):
> Measure of blood glucose levels ~ 2 hours after a meal.

> Elevated post-meal glucose is an early sign of impaired glucose metabolism and type 2 diabetes.

> Ranges: <18 µU/mL too Low; Typical peak: 18–276 mg/dL = Healty insulin response; ≥276 mg/dL = insulin resistance or hyperinsulinemia

- hba1c (float):
> Hemoglobin A1c

> Average blood sugar over 2–3 months.

> Diagnostic use: HbA1c 5.7–6.4% = prediabetes. HbA1c ≥6.5% = diabetes

> Strength: Does not require fasting and is widely used for diagnosis and monitoring.

- hdl_cholesterol (int):
> High-Density Lipoprotein Cholesterol: “Good” cholesterol. Removes excess cholesterol from arteries. Higher HDL = better.

> Ideal Level: >40 mg/dL (men), >50 mg/dL (women)

- heart_rate (int)
> Diabetes can cause autonomic neuropathy, leading to abnormal heart rate or variability.
> Not diagnostic, but may indicate complications.

- hypertension_history (int):
> Indicates whether or not the patient has a history of hypertension.

- income_level (string):
> The income bracket of the patient.

- insulin_level (float):
> Units: µU/mL (micro-units per milliliter) or mIU/L (milli-international units per liter). Units are interchangeable: 1 µU/mL = 1 mIU/L

> Fasting Insulin (after 8–12 hours without food):

> - Normal: 2–25 µU/mL
> - Optimal: Many experts suggest <10 µU/mL for best metabolic health
> - Above 25 µU/mL: May indicate insulin resistance or hyperinsulinemia
> - Below 2 µU/mL: May indicate type 1 diabetes or pancreatic dysfunction

> Postprandial (after eating, usually 30–90 minutes):

> - Normal peak: 18–276 µU/mL
> - Should return to baseline within 2–3 hours
> - Higher spikes often occur after high-carb meals

- ldl_cholesterol (int):
> Low-Density Lipoprotein Cholesterol: “Bad” cholesterol.

> Deposits cholesterol in artery walls, leading to plaque buildup. Lower LDL = better.

> Ideal Level: <100 mg/dL

- physical_activity_minutes_per_week (int):
> The average time of performing physical activities per week (in minutes).

- screen_time_hours_per_day (float):
> The time (in hours) that the patient spends using technology.

- sleep_hours_per_day (float):
> The amount of time the patient sleeps per night.

- smoking_status (string):
> The smoking status of the patient.

- systolic_bp (int):
> Normal peak: 18–276 µU/mL

> Should return to baseline within 2–3 hours

> Higher spikes often occur after high-carb meals

- triglycerides (int):
> A type of fat in blood. High levels often linked to insulin resistance and diabetes.

> Ideal Level: <150 mg/dL

- waist_to_hip_ratio (float):
> The waist circumference divided by the hip circumference.

- diagnosed_diabetes (int)
> Label/Target Column

# 📚 Data Sources
The data in this project is sourced from the following source: [Diabetes EDA Analysis](https://huggingface.co/datasets/guyshilo12/diabetes_eda_analysis)

许可证：MIT协议
任务类别：
- 令牌分类（Token Classification）
- 强化学习
语言：英语
标签：
- 健康
- 糖尿病
- 葡萄糖
- 分类
样本量范围：10K<n<100K

# ⛽ 合成糖尿病数据集
本数据集包含多种有助于预测患者是否罹患糖尿病的特征，已整合为单个CSV文件以供分析与模型训练。

# 📁 数据集说明
本数据集包含基于合成患者信息生成的模拟数据。

# 字段说明
本数据集包含以下字段：
- 腹部肥胖（abdominal_obesity，整数型）
- 每周酒精摄入量（alcohol_consumption_per_week，整数型）：患者每周的酒精摄入总量
- 酒精摄入分组（alcohol_group，字符串型）：根据患者酒精摄入情况划分的分组类别
- 体重指数（Body Mass Index, BMI，浮点型）：患者的体重指数
- BMI分组（bmi_group，字符串型）：根据患者体重指数划分的分组类别
- 心血管病史（cardiovascular_history，整数型）：标识患者是否存在心血管病史
- 总胆固醇（cholesterol_total，整数型）：血液中各类胆固醇的总和，是糖尿病性血脂异常的指标——该指标指糖尿病患者常见的异常血脂模式。总胆固醇=高密度脂蛋白胆固醇(High-Density Lipoprotein, HDL)+低密度脂蛋白胆固醇(Low-Density Lipoprotein, LDL)+极低密度脂蛋白胆固醇(Very Low-Density Lipoprotein, VLDL)=HDL+LDL+(甘油三酯/5)。理想水平：总胆固醇<200mg/dL
- 糖尿病风险评分（diabetes_risk_score，浮点型）
- 糖尿病分期（diabetes_stage，字符串型）
- 舒张压（diastolic_bp，整数型）：心脏舒张间歇期的血压。糖尿病患者常出现血压升高，但该指标无法用于糖尿病诊断，仅有助于风险分层
- 教育水平（education_level，字符串型）：患者的教育程度
- 就业状态（employment_status，字符串型）：患者的就业情况
- 种族（ethnicity，字符串型）：患者的种族背景
- 糖尿病家族史（family_history_diabetes，整数型）：标识患者家族是否存在糖尿病病史
- 性别（gender，字符串型）：患者的性别
- 空腹血糖（glucose_fasting，整数型）：患者空腹状态下的血糖水平
- 餐后血糖（glucose_postprandial，整数型）：餐后约2小时的血糖测量值。餐后血糖升高是葡萄糖代谢受损与2型糖尿病的早期征兆。参考范围：<18µU/mL为血糖过低；典型峰值18–276mg/dL为健康胰岛素反应；≥276mg/dL提示胰岛素抵抗或高胰岛素血症
- 糖化血红蛋白（Hemoglobin A1c, HbA1c，浮点型）：反映患者2-3个月的平均血糖水平。诊断标准：HbA1c 5.7–6.4%为糖尿病前期，HbA1c≥6.5%可确诊糖尿病。优势：无需空腹采血，广泛用于糖尿病的诊断与病情监测
- 高密度脂蛋白胆固醇（HDL-C，整数型）：又称“有益”胆固醇，可清除动脉内多余胆固醇，水平越高越好。理想水平：男性>40mg/dL，女性>50mg/dL
- 心率（heart_rate，整数型）：糖尿病可引发自主神经病变，导致心率或心率变异性异常，该指标无法用于糖尿病诊断，但可提示并发症风险
- 高血压病史（hypertension_history，整数型）：标识患者是否存在高血压病史
- 收入水平（income_level，字符串型）：患者所属的收入层级
- 胰岛素水平（insulin_level，浮点型）：单位：µU/mL（微单位/毫升）或mIU/L（毫国际单位/升），二者可互换：1µU/mL=1mIU/L。空腹胰岛素（禁食8-12小时后）：正常范围2–25µU/mL；多数专家建议<10µU/mL以获得最佳代谢健康；>25µU/mL可能提示胰岛素抵抗或高胰岛素血症；<2µU/mL可能提示1型糖尿病或胰腺功能异常。餐后胰岛素（进食后30-90分钟）：正常峰值18–276µU/mL，应在2-3小时内恢复至基线水平，高碳水饮食后常出现更高峰值
- 低密度脂蛋白胆固醇（LDL-C，整数型）：又称“有害”胆固醇，可将胆固醇沉积于动脉壁形成斑块，水平越低越好。理想水平：<100mg/dL
- 每周体育活动时长（physical_activity_minutes_per_week，整数型）：患者每周平均体育活动时间，单位为分钟
- 每日屏幕使用时长（screen_time_hours_per_day，浮点型）：患者每日使用电子设备的时长，单位为小时
- 每日睡眠时长（sleep_hours_per_day，浮点型）：患者每晚的睡眠时间
- 吸烟状态（smoking_status，字符串型）：患者的吸烟情况
- 收缩压（systolic_bp，整数型）：>正常峰值：18–276 µU/mL；>应在2-3小时内恢复至基线水平；>高碳水饮食后常出现更高峰值（注：原文此处存在复制粘贴错误，实际应为收缩压相关说明，按原文直译）
- 甘油三酯（triglycerides，整数型）：血液中的一类脂肪，高水平常与胰岛素抵抗和糖尿病相关，理想水平：<150mg/dL
- 腰臀比（waist_to_hip_ratio，浮点型）：腰围与臀围的比值
- 糖尿病确诊状态（diagnosed_diabetes，整数型）：标签/目标列

# 📚 数据来源
本项目使用的数据来源于以下数据集：[糖尿病探索性数据分析](https://huggingface.co/datasets/guyshilo12/diabetes_eda_analysis)