Predicting Goals in Hockey: An Expected Goals Model Using Machine Learning

import pandas as pd df_original = pd.read_csv("EH_pbp_query_20202021_2021-06-25.csv") df_original.head()

import numpy as np # get dataframe with shots only and make random sample of 5175 of the observations df_no_goals = df_original[df_original["event_type"] == "SHOT"] shot_rows_sample = np.random.choice(df_no_goals.index, 5176, False) shots_sample = df_no_goals.index.isin(shot_rows_sample) df_shots_sample = df_no_goals[shots_sample]

# get dataframe of goals only and append with shots-only sample dataframe df_only_goals = df_original[df_original["event_type"] == "GOAL"] df_all = df_only_goals.append(df_shots_sample) df_all.head()

df_all["event_type"].value_counts()

# drop columns with string data df_all = df_all.drop(["season", "game_date", "session", "clock_time", "event_description","event_team", "event_player_1", "home_on_1", "home_on_2", "home_on_3", "home_on_4", "home_on_5", "home_on_6","home_on_7", "away_on_1", "away_on_2", "away_on_3", "away_on_4", "away_on_5", "away_on_6","away_on_7", "home_goalie", "away_goalie", "home_team", "away_team", "game_score_state", "home_zone", "pred_goal"], axis=1) df_all.head()

df_event_type = df_all.event_type.str.get_dummies() df_event_type.head()

df_shot_type = df_all.event_detail.str.get_dummies() df_shot_type.head()

df_event_zone = df_all.event_zone.str.get_dummies() df_event_zone.head()

def if_assist(column): column = column.fillna("0") char = [] for value in column: if "." in value: char.append(value) for value in column: if value in char: column = column.replace(value, 1) column = pd.to_numeric(column) return column df_all["event_player_2"] = if_assist(df_all["event_player_2"])

df_all["event_player_3"] = if_assist(df_all["event_player_3"])

df_all["event_player_3"].value_counts()

df_all.head()

# put all tables together df_all = df_all.join(df_event_type).join(df_shot_type).join(df_event_zone) df_all.head()

# drop redundant columns df_all = df_all.drop(["event_type", "event_detail", "event_zone"], axis=1) df_all.head()

# final round of removing unecessary columns df_all = df_all.drop(columns=["game_id", "game_period", "game_seconds", "home_skaters", "away_skaters", "home_score", "away_score", "Def", "Neu", "Off", "event_index", "coords_x", "coords_y", "num_on", "num_off", "players_on", "players_off", "game_strength_state", "pbp_distance", "home_zonestart", "face_index", "pen_index", "shift_index", "SHOT"]) df_all.head()

# drop the rows with NaN values df_all = df_all.dropna(how="any") df_all.head()

# move "GOAL" column to first goal_col = "GOAL" first_col = df_all.pop(goal_col) df_all.insert(12, goal_col, first_col) df_all.head()

df_all.to_csv("hockey_data_1.csv", index=False)