ellamoses.github.io

Ella Moses and Kailen Mitchell

Project Title: NFL Injuries

Project Description¶

Our project works with a dataset titled "NFL 1st and Future - Analytics" which contains two files, Injury Record and Play List. The Injury Record file provides information about each injury including the player it happened to, the play and game it happened during, the surface it happened on, and the days missed due to injury. The Play List provides a list of each player’s plays during different games and other helpful information like the play type, temperature, stadium type, weather, and player’s position.

Link to the source of our data: https://www.kaggle.com/competitions/nfl-playing-surface-analytics/overview

Project Goals¶

Our goal is to use these two data sets to analyze the relationship between injuries in the NFL and other factors including the surface type of the field, the weather, player position, and play type. We will focus on predicting the body part that is injured and the days missed due to injury. Because football is an injury-prone sport, we would like to see if we can predict what types of injuries will occur so precautions can be taken to reduce the risk and severity.

Collaboration Plan

a) How are we working together?

  • We are working together by meeting up to work simultaneously on the project so that we can communicate effectively and divide tasks evenly. We will plan so that each of us will create an equal amount of visualizations and conclusions from the data so that both of us come away with a strong understanding of our data and how we found our results. It’s important to us that we are both working closely with the data as well as writing up results.

b) What technologies are we using?

  • We are using Google Collab/Jupiter Notebooks to parse the datasets and generate visualizations/results. At this time, we aren’t planning on using any other tools.

c) When / how often are we planning to meet?

  • We have very similar schedules and are in a lot of classes together so it’ll be very easy for us to find times to meet. Whenever we have a milestone coming up, we’ll probably meet 2-3 times per week during our break in between classes or after class if needed as well.

Discussion:¶

Can we use player position, weather, play type, and surface to predict the type and severity of an injury? Our data is useful because we can link the injuries to the plays and conditions the injury happened under.

Data Processing¶

We will begin by loading in our data and creating two dataframes to store the plays and injuries.

In [1]:
# clone the project repository, change to right directory, and import libraries.
%cd /content
!git clone https://github.com/EllaMoses/EllaMoses.github.io.git
%cd /content/EllaMoses.github.io/data
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns

/content
Cloning into 'EllaMoses.github.io'...
remote: Enumerating objects: 54, done.
remote: Counting objects: 100% (54/54), done.
remote: Compressing objects: 100% (47/47), done.
remote: Total 54 (delta 19), reused 3 (delta 1), pack-reused 0
Receiving objects: 100% (54/54), 2.06 MiB | 2.51 MiB/s, done.
Resolving deltas: 100% (19/19), done.
/content/EllaMoses.github.io/data
In [2]:
#Read in Play List file
plays = pd.read_csv("PlayList.csv")
plays
Out[2]:
PlayerKey GameID PlayKey RosterPosition PlayerDay PlayerGame StadiumType FieldType Temperature Weather PlayType PlayerGamePlay Position PositionGroup
0 26624 26624-1 26624-1-1 Quarterback 1 1 Outdoor Synthetic 63 Clear and warm Pass 1 QB QB
1 26624 26624-1 26624-1-2 Quarterback 1 1 Outdoor Synthetic 63 Clear and warm Pass 2 QB QB
2 26624 26624-1 26624-1-3 Quarterback 1 1 Outdoor Synthetic 63 Clear and warm Rush 3 QB QB
3 26624 26624-1 26624-1-4 Quarterback 1 1 Outdoor Synthetic 63 Clear and warm Rush 4 QB QB
4 26624 26624-1 26624-1-5 Quarterback 1 1 Outdoor Synthetic 63 Clear and warm Pass 5 QB QB
... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
267000 47888 47888-13 47888-13-51 Cornerback 99 13 Outdoor Synthetic 33 Sunny and cold Pass 51 DB DB
267001 47888 47888-13 47888-13-52 Cornerback 99 13 Outdoor Synthetic 33 Sunny and cold Pass 52 DB DB
267002 47888 47888-13 47888-13-53 Cornerback 99 13 Outdoor Synthetic 33 Sunny and cold Pass 53 DB DB
267003 47888 47888-13 47888-13-54 Cornerback 99 13 Outdoor Synthetic 33 Sunny and cold Pass 54 DB DB
267004 47888 47888-13 47888-13-55 Cornerback 99 13 Outdoor Synthetic 33 Sunny and cold Rush 55 DB DB

267005 rows × 14 columns

In [3]:
# data types of variables in plays dataframe
plays.dtypes
Out[3]:
PlayerKey          int64
GameID            object
PlayKey           object
RosterPosition    object
PlayerDay          int64
PlayerGame         int64
StadiumType       object
FieldType         object
Temperature        int64
Weather           object
PlayType          object
PlayerGamePlay     int64
Position          object
PositionGroup     object
dtype: object
In [4]:
# summary statistics for plays dataframe
plays.describe()
Out[4]:
PlayerKey PlayerDay PlayerGame Temperature PlayerGamePlay
count 267005.000000 267005.000000 267005.000000 267005.000000 267005.000000
mean 41515.381465 210.451351 13.799131 -35.029535 29.058647
std 4125.858924 183.643654 8.342894 304.583110 19.626551
min 26624.000000 -62.000000 1.000000 -999.000000 1.000000
25% 39653.000000 43.000000 7.000000 44.000000 13.000000
50% 42432.000000 102.000000 13.000000 61.000000 26.000000
75% 44480.000000 400.000000 20.000000 72.000000 43.000000
max 47888.000000 480.000000 32.000000 97.000000 102.000000
In [5]:
#Read in injury file
injury = pd.read_csv("InjuryRecord.csv")
injury
Out[5]:
PlayerKey GameID PlayKey BodyPart Surface DM_M1 DM_M7 DM_M28 DM_M42
0 39873 39873-4 39873-4-32 Knee Synthetic 1 1 1 1
1 46074 46074-7 46074-7-26 Knee Natural 1 1 0 0
2 36557 36557-1 36557-1-70 Ankle Synthetic 1 1 1 1
3 46646 46646-3 46646-3-30 Ankle Natural 1 0 0 0
4 43532 43532-5 43532-5-69 Ankle Synthetic 1 1 1 1
... ... ... ... ... ... ... ... ... ...
100 44423 44423-13 NaN Knee Synthetic 1 0 0 0
101 31933 31933-20 NaN Knee Synthetic 1 0 0 0
102 47285 47285-4 NaN Knee Natural 1 1 0 0
103 37068 37068-19 NaN Knee Natural 1 1 0 0
104 36696 36696-24 NaN Knee Synthetic 1 1 0 0

105 rows × 9 columns

In [6]:
# data types of variables in injury dataframe
injury.dtypes
Out[6]:
PlayerKey     int64
GameID       object
PlayKey      object
BodyPart     object
Surface      object
DM_M1         int64
DM_M7         int64
DM_M28        int64
DM_M42        int64
dtype: object
In [7]:
# summary statistics for injury data frame
injury.describe()
Out[7]:
PlayerKey DM_M1 DM_M7 DM_M28 DM_M42
count 105.000000 105.0 105.000000 105.000000 105.000000
mean 42283.609524 1.0 0.723810 0.352381 0.276190
std 4163.510366 0.0 0.449257 0.480003 0.449257
min 31070.000000 1.0 0.000000 0.000000 0.000000
25% 39656.000000 1.0 0.000000 0.000000 0.000000
50% 43518.000000 1.0 1.000000 0.000000 0.000000
75% 45966.000000 1.0 1.000000 1.000000 1.000000
max 47813.000000 1.0 1.000000 1.000000 1.000000

We will now combine the two dataframes by performing a merge to add all relevent information from plays data frame to the injury data frame. We will then replace our missing values with NaN and clean our weather column values. The four DM_M variables are dummy variables used to encode whether or not a player missed, 1, 7, 28, or 42 days. We will create a new column to store the days missed so we are able to work with the actual number of days in addition to the provided binary variables.

In [8]:
#Combine the two dataframes by performing a right merge to add all relevent info from plays data frame to the injury data frame

injury_plays = injury.merge(plays, on=["PlayerKey", "GameID", "PlayKey"], how="left")

#we can use .replace to fix all -999 values with NaN
injury_plays['Temperature'] = injury_plays['Temperature'].replace(-999,np.nan)

# map weather conditions to account for typos and other name variations
weathermapped = injury_plays['Weather'].map({
  "Clear": "Clear",
  "Clear Skies": "Clear",
  "Clear and warm":"Clear",
  "Clear skies":"Clear",
  "Cloudy":"Cloudy",
  "Cloudy and Cool":"Cloudy",
  "Cloudy with periods of rain, thunder possible. Winds shifting to WNW, 10-20 mph.":"Rain",
  "Cloudy, 50% chnace of rain":"Rain",
  "Cold":"NaN",
  "Controlled Climate":"Indoor",
  "Coudy":"Cloudy",
  "Fair":"Partly Cloudy",
  "Indoor":"Indoor",
  "Indoors":"Indoor",
  "Light Rain":"Rain",
  "Mostly Sunny":"Partly Cloudy",
  "Mostly cloudy":"Partly Cloudy",
  "Mostly sunny":"Partly Cloudy",
  "Partly Cloudy":"Partly Cloudy",
  "Rain":"Rain",
  "Rain shower":"Rain",
  "Sun & clouds":"Partly Cloudy",
  "Sunny":"Clear",
})
injury_plays['Weather'] = weathermapped

# Create a new variable to store the number of days missed
def num_days_missed(row):
    for i in [42, 28, 7, 1]:
        if row[f'DM_M{i}'] == 1:
            return i
    return 0

# Apply the function to each row to get the total days missed
injury_plays['total_days_missed'] = injury_plays.apply(num_days_missed, axis=1)

injury_plays
Out[8]:
PlayerKey GameID PlayKey BodyPart Surface DM_M1 DM_M7 DM_M28 DM_M42 RosterPosition ... PlayerGame StadiumType FieldType Temperature Weather PlayType PlayerGamePlay Position PositionGroup total_days_missed
0 39873 39873-4 39873-4-32 Knee Synthetic 1 1 1 1 Linebacker ... 4.0 Indoors Synthetic 84.0 Cloudy Punt 32.0 OLB LB 42
1 46074 46074-7 46074-7-26 Knee Natural 1 1 0 0 Linebacker ... 7.0 Open Natural 76.0 Partly Cloudy Punt 26.0 OLB LB 7
2 36557 36557-1 36557-1-70 Ankle Synthetic 1 1 1 1 Safety ... 1.0 Outdoor Synthetic 63.0 Clear Pass 70.0 SS DB 42
3 46646 46646-3 46646-3-30 Ankle Natural 1 0 0 0 Linebacker ... 3.0 Outdoor Natural 80.0 Cloudy Punt 30.0 LB LB 1
4 43532 43532-5 43532-5-69 Ankle Synthetic 1 1 1 1 Wide Receiver ... 5.0 Retractable Roof Synthetic 89.0 Partly Cloudy Kickoff 69.0 WR WR 42
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
100 44423 44423-13 NaN Knee Synthetic 1 0 0 0 NaN ... NaN NaN NaN NaN NaN NaN NaN NaN NaN 1
101 31933 31933-20 NaN Knee Synthetic 1 0 0 0 NaN ... NaN NaN NaN NaN NaN NaN NaN NaN NaN 1
102 47285 47285-4 NaN Knee Natural 1 1 0 0 NaN ... NaN NaN NaN NaN NaN NaN NaN NaN NaN 7
103 37068 37068-19 NaN Knee Natural 1 1 0 0 NaN ... NaN NaN NaN NaN NaN NaN NaN NaN NaN 7
104 36696 36696-24 NaN Knee Synthetic 1 1 0 0 NaN ... NaN NaN NaN NaN NaN NaN NaN NaN NaN 7

105 rows × 21 columns

Exploratory Analysis & Data Visualization¶

We will now create tables and visualziations of our data to analyze the relationships between our variables. We will focus on the frequencies of types and severities of injuries based on Weather, Surface Type, Play Type, and position.

In [ ]:
bodypartgroup = injury.groupby('BodyPart').count()
bodypartgroup['PlayerKey'].plot.bar()
Out[ ]:
<Axes: xlabel='BodyPart'>

The most common injuries are knee and ankle injuries.

In [ ]:
injury_plays['total_days_missed'].value_counts().sort_index().plot.bar()
Out[ ]:
<Axes: xlabel='total_days_missed'>

It is most common for a player to miss seven days due to injury.

In [ ]:
pd.crosstab(index=injury_plays["total_days_missed"],
                        columns=injury_plays["BodyPart"],normalize=True)
Out[ ]:
BodyPart Ankle Foot Heel Knee Toes
total_days_missed
1 0.152381 0.000000 0.000000 0.104762 0.019048
7 0.123810 0.000000 0.009524 0.200000 0.038095
28 0.019048 0.019048 0.000000 0.028571 0.009524
42 0.104762 0.047619 0.000000 0.123810 0.000000

Above table shows the frequency of days missed based on body part injured.

In [ ]:
sns.heatmap( pd.crosstab(injury_plays['total_days_missed'], injury_plays['BodyPart'], normalize=True), cmap='Purples')
Out[ ]:
<Axes: xlabel='BodyPart', ylabel='total_days_missed'>

This visualization shows the frequency of total days missed based on the type of injury.

In [ ]:
pd.crosstab(index=injury_plays["Weather"],
                        columns=injury_plays["BodyPart"])
Out[ ]:
BodyPart Ankle Foot Knee
Weather
Clear 13 0 9
Cloudy 6 2 7
Indoor 4 0 5
NaN 1 1 1
Partly Cloudy 8 1 7
Rain 2 0 5

Above table shows the frequency of injury type of injury based on weather.

In [ ]:
sns.heatmap( pd.crosstab(injury_plays['BodyPart'], injury_plays['Weather'], normalize=True), cmap='Purples')
Out[ ]:
<Axes: xlabel='Weather', ylabel='BodyPart'>

This visualization shows the frequency of different types of injuries based on the weather.

In [ ]:
pd.crosstab(index=injury_plays["Weather"],
                        columns=injury_plays["total_days_missed"])
Out[ ]:
total_days_missed 1 7 28 42
Weather
Clear 4 8 0 10
Cloudy 3 6 1 5
Indoor 3 4 0 2
NaN 1 1 0 1
Partly Cloudy 3 4 5 4
Rain 2 4 0 1

Above table shows the frequency of days missed based on weather.

In [ ]:
sns.heatmap( pd.crosstab(injury_plays['total_days_missed'], injury_plays['Weather'], normalize=True), cmap='Purples')
Out[ ]:
<Axes: xlabel='Weather', ylabel='total_days_missed'>

This visualization shows the frequency of days missed based on the weather.

In [ ]:
pd.crosstab(index=injury_plays["Surface"],
                        columns=injury_plays["BodyPart"],normalize=True)
Out[ ]:
BodyPart Ankle Foot Heel Knee Toes
Surface
Natural 0.161905 0.047619 0.009524 0.228571 0.009524
Synthetic 0.238095 0.019048 0.000000 0.228571 0.057143

Above table shows the frequency of injury type based on surface type.

In [ ]:
sns.heatmap( pd.crosstab(injury_plays['BodyPart'], injury_plays['Surface'],normalize=True), cmap='Purples')
Out[ ]:
<Axes: xlabel='Surface', ylabel='BodyPart'>

The heatmap above shows the frequency of injury for each body part based on the suface. Knee injuries are the most frequent and the number does not appear to be affected by surface type.

In [ ]:
pd.crosstab(index=injury_plays["Surface"],
                        columns=injury_plays["total_days_missed"])
Out[ ]:
total_days_missed 1 7 28 42
Surface
Natural 13 20 2 13
Synthetic 16 19 6 16

Above table shows the frequency of days missed based on surface.

In [ ]:
sns.heatmap( pd.crosstab(injury_plays['total_days_missed'], injury_plays['Surface'], normalize=True), cmap='Purples')
Out[ ]:
<Axes: xlabel='Surface', ylabel='total_days_missed'>

This visualization shows the frequency of days missed based on the surface.

In [ ]:
pd.crosstab(index=injury_plays["PlayType"],
                        columns=injury_plays["BodyPart"], normalize=True)
Out[ ]:
BodyPart Ankle Foot Knee
PlayType
Kickoff 0.012987 0.012987 0.064935
Kickoff Not Returned 0.000000 0.000000 0.012987
Kickoff Returned 0.000000 0.000000 0.012987
Pass 0.194805 0.038961 0.181818
Punt 0.038961 0.000000 0.077922
Punt Not Returned 0.000000 0.012987 0.000000
Punt Returned 0.025974 0.000000 0.012987
Rush 0.181818 0.012987 0.103896

Above table shows the frequency of injury type based on play type.

In [ ]:
sns.heatmap( pd.crosstab(injury_plays['BodyPart'], injury_plays['PlayType'], normalize=True), cmap='Purples')
Out[ ]:
<Axes: xlabel='PlayType', ylabel='BodyPart'>

This visualization shows the frequency of different types of injuries based on the play type.

In [ ]:
pd.crosstab(index=injury_plays["PlayType"],
                        columns=injury_plays["total_days_missed"])
Out[ ]:
total_days_missed 1 7 28 42
PlayType
Kickoff 1 2 0 4
Kickoff Not Returned 0 1 0 0
Kickoff Returned 0 1 0 0
Pass 7 13 3 9
Punt 2 4 0 3
Punt Not Returned 0 0 1 0
Punt Returned 0 2 0 1
Rush 7 6 3 7

Above table shows the frequency of days missed based on play type.

In [ ]:
sns.heatmap( pd.crosstab(injury_plays['total_days_missed'], injury_plays['PlayType'], normalize=True), cmap='Purples')
Out[ ]:
<Axes: xlabel='PlayType', ylabel='total_days_missed'>

This visualization shows the frequency of days missed based on the play type.

In [ ]:
pd.crosstab(index=injury_plays["Position"],
                        columns=injury_plays["BodyPart"],normalize=True)
Out[ ]:
BodyPart Ankle Foot Knee
Position
C 0.038961 0.012987 0.000000
CB 0.064935 0.012987 0.025974
DB 0.000000 0.000000 0.012987
DE 0.025974 0.012987 0.025974
DT 0.000000 0.000000 0.025974
FS 0.038961 0.000000 0.025974
ILB 0.012987 0.000000 0.025974
LB 0.025974 0.000000 0.000000
MLB 0.038961 0.000000 0.012987
OLB 0.051948 0.000000 0.103896
RB 0.025974 0.000000 0.051948
SS 0.025974 0.000000 0.038961
T 0.000000 0.012987 0.012987
TE 0.000000 0.012987 0.012987
WR 0.103896 0.012987 0.090909

Above table shows the frequency of injury type based on position.

In [ ]:
sns.heatmap( pd.crosstab(injury_plays['BodyPart'], injury_plays['Position'],normalize=True), cmap='Purples')
Out[ ]:
<Axes: xlabel='Position', ylabel='BodyPart'>

This visualization shows the frequency of different types of injuries based on the Position. We can see that WR tend to have the most injuries. Ankle injuries appear to be the most common.

In [ ]:
pd.crosstab(index=injury_plays["Position"],
                        columns=injury_plays["total_days_missed"],normalize=True)
Out[ ]:
total_days_missed 1 7 28 42
Position
C 0.025974 0.012987 0.000000 0.012987
CB 0.012987 0.051948 0.025974 0.012987
DB 0.000000 0.012987 0.000000 0.000000
DE 0.000000 0.038961 0.000000 0.025974
DT 0.000000 0.012987 0.000000 0.012987
FS 0.012987 0.038961 0.012987 0.000000
ILB 0.012987 0.012987 0.000000 0.012987
LB 0.025974 0.000000 0.000000 0.000000
MLB 0.012987 0.012987 0.012987 0.012987
OLB 0.025974 0.064935 0.012987 0.051948
RB 0.000000 0.025974 0.000000 0.051948
SS 0.012987 0.000000 0.000000 0.051948
T 0.000000 0.012987 0.000000 0.012987
TE 0.000000 0.012987 0.012987 0.000000
WR 0.077922 0.064935 0.012987 0.051948

Frequencies of total days missed based on position.

In [ ]:
sns.heatmap( pd.crosstab(injury_plays['total_days_missed'], injury_plays['Position'],normalize=True), cmap='Purples')
Out[ ]:
<Axes: xlabel='Position', ylabel='total_days_missed'>

This visualization shows the frequency of days missed based on the position. We can see that WR tend to have the most severe injuries.

Based on the above visualizations, we can see that there are some trends in our data such as higher injury counts for some positions and more severe injuries in certain weather conditions. The trends do not appear to be strong, so we are expecting that while these four variables may help us predict injury types and severities, our accuracy will not be very high.

Model¶

We plan to use conditions such as player position, weather, play type, and field type to determine the type and severity of an injury. We will use K-nearest neighbors classification on the below dataset for this model. We will also compare the KNN results to a logistic regression model.

In [11]:
import warnings
warnings.filterwarnings('ignore')
injury_plays = injury_plays.dropna()
injury_plays["string_days_missed"] = injury_plays["total_days_missed"].astype(str)
injury_plays
Out[11]:
PlayerKey GameID PlayKey BodyPart Surface DM_M1 DM_M7 DM_M28 DM_M42 RosterPosition ... StadiumType FieldType Temperature Weather PlayType PlayerGamePlay Position PositionGroup total_days_missed string_days_missed
0 39873 39873-4 39873-4-32 Knee Synthetic 1 1 1 1 Linebacker ... Indoors Synthetic 84.0 Cloudy Punt 32.0 OLB LB 42 42
1 46074 46074-7 46074-7-26 Knee Natural 1 1 0 0 Linebacker ... Open Natural 76.0 Partly Cloudy Punt 26.0 OLB LB 7 7
2 36557 36557-1 36557-1-70 Ankle Synthetic 1 1 1 1 Safety ... Outdoor Synthetic 63.0 Clear Pass 70.0 SS DB 42 42
3 46646 46646-3 46646-3-30 Ankle Natural 1 0 0 0 Linebacker ... Outdoor Natural 80.0 Cloudy Punt 30.0 LB LB 1 1
4 43532 43532-5 43532-5-69 Ankle Synthetic 1 1 1 1 Wide Receiver ... Retractable Roof Synthetic 89.0 Partly Cloudy Kickoff 69.0 WR WR 42 42
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
90 42418 42418-19 42418-19-15 Ankle Synthetic 1 0 0 0 Offensive Lineman ... Retr. Roof - Closed Synthetic 57.0 Partly Cloudy Rush 15.0 C OL 1 1
91 46394 46394-18 46394-18-3 Knee Synthetic 1 1 0 0 Tight End ... Outdoor Synthetic 45.0 Cloudy Kickoff Returned 3.0 TE TE 7 7
92 45187 45187-9 45187-9-4 Ankle Natural 1 0 0 0 Wide Receiver ... Outdoor Natural 81.0 Cloudy Rush 4.0 WR WR 1 1
93 42448 42448-14 42448-14-3 Knee Synthetic 1 1 1 0 Wide Receiver ... Retractable Roof Synthetic 78.0 Partly Cloudy Pass 3.0 WR WR 28 28
94 47334 47334-8 47334-8-1 Knee Synthetic 1 1 0 0 Safety ... Indoor Synthetic 74.0 Clear Kickoff Not Returned 1.0 DB DB 7 7

63 rows × 22 columns

For our KNN classification models we will define our training data and select the number of neighbors and cross validation folds to maximize our accuracy. For our logistic regression models we will also try to select the number of cross validation folds that maximizes our accuracy.

In [13]:
# KNN to predict body part
from sklearn.preprocessing import StandardScaler
from sklearn.neighbors import KNeighborsClassifier
from sklearn.feature_extraction import DictVectorizer
from sklearn.model_selection import cross_val_score
from sklearn.metrics import make_scorer, recall_score, precision_score
from sklearn.pipeline import Pipeline

# define the training data
X_train = injury_plays[["Weather", "Position", "PlayType", "FieldType"]].to_dict(orient="records")
y_train = injury_plays["BodyPart"]

#need to do dummy data here
vec = DictVectorizer(sparse=False)
vec.fit(X_train)
X_train = vec.transform(X_train)


# standardize the data
scaler = StandardScaler()
scaler.fit(X_train)
X_train_sc = scaler.transform(X_train)

# fit the k-nearest neighbors model
model = KNeighborsClassifier(n_neighbors=7)
model.fit(X_train_sc, y_train)

pipeline = Pipeline([
    ("scaler", scaler),
    ("model", model)
])
cross_val_score(pipeline, X_train, y_train,
                cv=3, scoring="accuracy").mean()
Out[13]:
0.5396825396825397

When using a KNN classification model to predict body part based on weather, position, play type, and field type, the accuracy is about 0.54.

In [14]:
# logistic regression to predict body part
from sklearn.linear_model import LogisticRegression

# define the training data
X_train = injury_plays[["Weather", "Position", "PlayType", "FieldType"]].to_dict(orient="records")
y_train = injury_plays["BodyPart"]

#need to do dummy data here
vec = DictVectorizer(sparse=False)
vec.fit(X_train)
X_train = vec.transform(X_train)


# standardize the data
scaler = StandardScaler()
scaler.fit(X_train)
X_train_sc = scaler.transform(X_train)

# fit a logistic regression model
model = LogisticRegression()
model.fit(X_train_sc, y_train)
pipeline = Pipeline([
    ("scaler", scaler),
    ("model", model)
])
cross_val_score(pipeline, X_train, y_train,
                cv=3, scoring="accuracy").mean()
Out[14]:
0.5396825396825397

When using a logisitic regression model to predict body part based on weather, position, play type, and field type, the accuracy is about 0.54. This is the same accuracy as the KNN model.

In [15]:
# KNN for days missed
# define the training data
X_train = injury_plays[["Weather", "Position", "PlayType", "FieldType"]].to_dict(orient="records")
y_train = injury_plays["string_days_missed"]

#need to do dummy data here
vec = DictVectorizer(sparse=False)
vec.fit(X_train)
X_train = vec.transform(X_train)


# standardize the data
scaler = StandardScaler()
scaler.fit(X_train)
X_train_sc = scaler.transform(X_train)

# fit the k-nearest neighbors model, using classifier because string_days_missed is an object
model = KNeighborsClassifier(n_neighbors=10)
model.fit(X_train_sc, y_train)
pipeline = Pipeline([
    ("scaler", scaler),
    ("model", model)
])
cross_val_score(pipeline, X_train, y_train,
                cv=5, scoring="accuracy").mean()
Out[15]:
0.43076923076923085

When using a KNN classification model to predict days missed based on weather, position, play type, and field type, the accuracy is about 0.43.

In [16]:
# logistic regression for days missed

# define the training data
X_train = injury_plays[["Weather", "Position", "PlayType", "FieldType"]].to_dict(orient="records")
y_train = injury_plays["string_days_missed"]

#need to do dummy data here
vec = DictVectorizer(sparse=False)
vec.fit(X_train)
X_train = vec.transform(X_train)


# standardize the data
scaler = StandardScaler()
scaler.fit(X_train)
X_train_sc = scaler.transform(X_train)

# fit a logistic model
model = LogisticRegression()
model.fit(X_train_sc, y_train)
pipeline = Pipeline([
    ("scaler", scaler),
    ("model", model)
])
cross_val_score(pipeline, X_train, y_train,
                cv=8, scoring="accuracy").mean()
Out[16]:
0.4419642857142857

When using a logistic regression model to predict days missed based on weather, position, play type, and field type, the accuracy is about 0.44. This is approximately the same accuracy as the KNN model, but slightly higher.

Conclusions¶

In conclusion, there is some correlaton between position, play type, weather, and field type with body part and severity but there is not enough of a correlation to consistently predict these traits. Our KNN and logistic regression models had almost identical accuracies. This leads us to the conclusion that NFL injuries are somewhat random, which makes them difficult to predict and prevent.

In future studies, we would be interested to see if it is possible to predict whether or not an injury occurs given the above variables and other factors related to a player's health and physical condition.

In [ ]:
# convert notebook to html file, must first download this notebook and add it to the content folder
%%shell
jupyter nbconvert --to html /content/KailenElla.ipynb