MLB-Win-Prediction

Predict MLB wins per season based on past data by regression model with some analyses

source : https://www.datacamp.com/community/tutorials/scikit-learn-tutorial-baseball-1

Prerequisites

• scikit-learn
• pandas
• sqlite3
• matplotlib

Workflow

• Import data from database to df(DataFrame) with pandas
• data pre-processing - add column headers to df, dealing with null values
• exploring and visualizing the data - distribution of wins by hist, show scatter plot from quantized data , figure about runs per-game with years
• feature extracting - one-hot encoding(get_dummies) according to yearID, runs per-game,runs allowed per-game
• Build the predicition model - add clustering information by K-means as feature, split data to train and test , build model by regression , select MAE(mean absolute error ) as error metric

Importing data

Import data from database to df(DataFrame) with pandas

• Each row of our data will consist of a single team for a specific year data from https://github.com/jknecht/baseball-archive-sqlite

+ Read data by quering a sqlitedatabase using the sqlite3 package
+ Convert th fetch data to a DataFrame with pandas
the data will be filtered to only include currently active teams and only years where the team played 150 or more games

Data pre-processing

Add column headers to dataframe , remove columns which aren't associated to the target column 'W' and deal with null values

Add column headers

Passing a list of headers to the columns attribeute from pandas to add headers

# names for the columns of dataframe
cols = ['yearID','lgID','teamID','franchID','divID','Rank','G','Ghome','W','L','DivWin','WCWin','LgWin','WSWin','R','AB','H','2B','3B','HR','BB','SO','SB','CS','HBP','SF','RA','ER','ERA','CG','SHO','SV','IPouts','HA','HRA','BBA','SOA','E','DP','FP','name','park','attendance','BPF','PPF','teamIDBR','teamIDlahman45','teamIDretro','franchID','franchName','active','NAassoc']

teams_df.columns = cols

Remove not necessary columns with respect to 'Win'

# Dropping unnecesary column variables.
drop_cols = ['lgID','franchID','divID','Rank','Ghome','L','DivWin','WCWin','LgWin','WSWin','SF','name','park','attendance','BPF','PPF','teamIDBR','teamIDlahman45','teamIDretro','franchID','franchName','active','NAassoc']

df = teams_df.drop(drop_cols, axis=1)

Deal with null values

Display the count of null values for each column

# Print out null values of all columns of `df`
print(df.isnull().sum(axis=0).tolist())

set SO(110 null values) and DP(22) to median of the column

(because the amount of null values to both columns are relatively small)

filling the null values by the fillna() method

e.g.
df['SO'] = df['SO'].fillna(df['SO'].median())
df['DP'] = df['DP'].fillna(df['DP'].median())

remove/drop CS(419 null values) and HBP(1777) columns

(Two of the columns have relatively high amount of null values)

df = df.drop(['CS','HBP'], axis=1)

Exploring and Visualizing the data

• histogram of the win distribution
• scatter plot to show the quantized data according to # of wins
• figure about average runs per-game with respect to years

histogram of the win distribution with matplotlib

e.g.
plt.hist(df['W'])
plt.xlabel('Wins')
plt.title('Distribution of Wins')

plt.show()

scatter plot to show the quantized data according to # of wins

create a new column win_bins by apply() method on the column 'W' and function 'assign_win_bins()' assign_win_bins : quantized the column 'W' from integer value to 1~5 bins

df['win_bins'] = df['W'].apply(assign_win_bins)

after extracting the column 'win_bins', make a scatter graph :

plt.scatter(df['yearID'], df['W'], c=df['win_bins'])
plt.title('Wins Scatter Plot')
plt.xlabel('Year')
plt.ylabel('Wins')

plt.show()

we can see in the above scatter plot, there are very few seasons from before 1900 because of that, it makes sense to eliminate those rows from data sets

df = df[ df['yearID'] > 1900 ] 

figure about average runs per-game with years to indicate how much scoring there was for each year

As MLB progressed, different eras emerged where the amount of runs per game increased or decreased significantly. The dead ball era of the early 1900s is an example of a low scoring era and the steroid era at the turn of the 21st century is an example of a high scoring era.

steps:

1. create dictionaries runs-per-year and games-per-year by using the 'iterrows()' method to loop through dataframe
2. create a dictionary called mlb-runs-per-game by iterate through the games-per-year with 'items()' method
3. sort mlb-runs-per-game and save as list befor unzip it

plot from the mlb-runs-per-game dictionary :

plt.plot(x, y)
plt.title('MLB Yearly Runs per Game')
plt.xlabel('Year')
plt.ylabel('MLB Runs per Game')

plt.show()

Adding Features

• era As we can see on above plot, the scoring trends are quite different with respect to era So we can create new feature that indicate a specifi era for each row of data based on the 'yearID' (solved with get_dummies/one-hot encoding

• run per game of specified year add run per game information for each row by using 'yearID' and function 'apply()'

• run per game for each team

• run allowed per game for each team

we can show the scatter plot about RPG and RAPG vs win

• labels derived from a K-means clustering algorithm provided by sklearn create a dataframe for clustering which leaves out the target variable('W')

attributes = ['G','R','AB','H','2B','3B','HR','BB','SO','SB','RA','ER','ERA','CG',
'SHO','SV','IPouts','HA','HRA','BBA','SOA','E','DP','FP','era_1','era_2','era_3','era_4','era_5','era_6','era_7','era_8','decade_1910','decade_1920','decade_1930','decade_1940','decade_1950','decade_1960','decade_1970','decade_1980','decade_1990','decade_2000','decade_2010','R_per_game','RA_per_game','mlb_rpg']

data_attributes = df[attributes]

Use sklearn’s 'silhouette_score()' function to determine how many clusters we want (This function returns the mean silhouette coefficient over all samples. You want a higher silhouette score, and the score decreases as more clusters are added)

Execute K-means model : set the number of cluster to 6 random state to 1 fit_transform() : determine the Euclidian distances for each data point

visualized the clusters with a scatter plot

before getting into any machine learing models...

we can use corr() method from Pandas to see how each variables/features is correlated with target variable

df.corr()['W']

Build Model

pre-processing

• create new dataframe using only variables to be included in models

numeric_cols = ['G','R','AB','H','2B','3B','HR','BB','SO','SB','RA','ER','ERA','CG','SHO','SV','IPouts','HA','HRA','BBA','SOA','E','DP','FP','era_1','era_2','era_3','era_4','era_5','era_6','era_7','era_8','decade_1910','decade_1920','decade_1930','decade_1940','decade_1950','decade_1960','decade_1970','decade_1980','decade_1990','decade_2000','decade_2010','R_per_game','RA_per_game','mlb_rpg','labels','W']

data = df[numeric_cols]

• split data to training and testing data

train = data.sample(frac=0.75, random_state=1)
test = data.loc[~data.index.isin(train.index)]

x_train = train[attributes]
y_train = train['W']
x_test = test[attributes]
y_test = test['W']

Select model and error metric

• linear/Ridge regression(Linear regression + regularizer)
• mean absolute error

steps:

1. import LinearRegression/RidgeCV and mean_absolute_error from sklearn.linear_model and sklearn.metrics respectively
2. create a model lr/rrm and fit the model
3. make predictions and determinie MAE of the model