nfl-4th-downs-dataset
收藏资源简介：
NFL四档进攻决策数据集（1999-2025）是一个专门用于分析和预测美式橄榄球比赛中教练在四档进攻时决策的机器学习数据集。该数据集从1999年至2025年共27个NFL赛季的完整比赛逐次播放数据中筛选出所有四档进攻场景，总计113,156次四档进攻播放，覆盖7,273场比赛。数据集的核心任务是预测教练在四档进攻时的决策类型，目标变量为三类：go（跑动或传球进攻）、punt（弃踢）和fg（射门尝试）。数据集中包含28个特征字段，涵盖比赛背景（赛季、比赛ID、节次、剩余时间等）、场地位置（距离对方端区距离、需要推进码数等）、球队信息（控球方、防守方、主客队等）、胜率预期和博彩背景（胜率、让分盘、总分盘等）。数据集按训练集（90,524个样本）、验证集（11,316个样本）和测试集（11,316个样本）划分，类别分布显示弃踢决策占主导（61.75%），射门尝试占23.77%，进攻尝试占14.49%。该数据集适用于构建分类模型，研究教练决策模式，分析比赛策略，以及体育分析相关的机器学习任务。
The NFL Fourth-Down Decision Dataset (1999-2025) is a specialized machine learning dataset designed for analyzing and predicting coaches decisions during fourth-down plays in American football games. It is derived from complete play-by-play data across 27 NFL seasons from 1999 to 2025, filtering all fourth-down scenarios, totaling 113,156 fourth-down plays covering 7,273 games. The core task of the dataset is to predict the type of coach decision on fourth down, with the target variable categorized into three classes: go (run or pass offensive attempt), punt (punt), and fg (field goal attempt). The dataset includes 28 feature fields covering game context (season, game ID, quarter, time remaining, etc.), field position (distance to opponents end zone, yards to gain, etc.), team information (possessing team, defending team, home/away team, etc.), win probability expectations, and betting background (win probability, point spread, over/under, etc.). The dataset is split into training set (90,524 samples), validation set (11,316 samples), and test set (11,316 samples). The class distribution shows that punt decisions dominate (61.75%), field goal attempts account for 23.77%, and offensive attempts account for 14.49%. This dataset is suitable for building classification models, studying coach decision patterns, analyzing game strategies, and machine learning tasks related to sports analytics.
数据集概述
数据集名称: NFL 4th Down Decision Dataset (1999–2025)
数据集地址: https://huggingface.co/datasets/afzalmengal/nfl-4th-downs-dataset
核心任务: 基于比赛上下文，预测NFL球队在第4档进攻时的教练决策（强攻、弃踢或射门）。
目标变量
label 字段表示决策类别，由原始 play_type 映射而来：
| 标签 | 含义 |
|---|---|
| go | 强攻（跑球或传球） |
| punt | 弃踢 |
| fg | 射门 |
特征说明
比赛上下文 (Game Context)
game_id: 唯一比赛标识season: NFL赛季（1999–2025）qtr: 节次game_seconds_remaining: 比赛剩余秒数score_differential: 持球队得分与防守队得分之差home_score: 主队得分away_score: 客队得分
场上位置 (Field Position)
yardline_100: 距对方端区的距离（0–100）ydstogo: 首攻所需码数goal_to_go: 是否处于达阵区前（1/0）
球队信息 (Teams)
posteam: 持球进攻队defteam: 防守队home_team: 主队away_team: 客队
胜率与博彩 (Win Expectancy / Betting)
wp: 胜率概率spread_line: 拉斯维加斯让分盘口total_line: 拉斯维加斯总分线
原始标签与标签
play_type: NFL原始进攻类型label: 清洗后的决策类别
数据集规模与统计
| 项目 | 数值 |
|---|---|
| 总行数（第4档进攻） | 113,156 |
| 总比赛数 | 7,273 |
| 覆盖赛季 | 27个（1999–2025） |
| 原始列数 | 372 |
类别分布（目标变量 label）
| 类别 | 数量 | 百分比 |
|---|---|---|
| go | 15,515 | 14.49% |
| punt | 66,137 | 61.75% |
| fg | 25,455 | 23.77% |
每场比赛平均第4档进攻次数: 15
每赛季平均比赛数: 269
数据集划分
| 划分 | 样本数 |
|---|---|
| 训练集 (train) | 90,524 |
| 验证集 (validation) | 11,316 |
| 测试集 (test) | 11,316 |
使用示例
python from datasets import load_dataset
dataset = load_dataset("your-username/nfl-fourth-down-dataset")
train_data = dataset["train"] print(train_data[0])
