# import libraries import pandas as pd import numpy as np import statsmodels.api as sm from statsmodels.formula.api import ols from statsmodels.stats.outliers_influence import variance_inflation_factor from sklearn.linear_model import LinearRegression from sklearn.model_selection import train_test_split from sklearn.metrics import r2_score, mean_squared_error import matplotlib.pyplot as plt import seaborn as sns sns.set_theme(style="darkgrid") sns.set_palette("deep") # load data df = pd.read_csv("twitch-streamers.csv") df.head()

# check data types df.dtypes

# convert Partnered and Mature columns df['Partnered'] = np.where(df['Partnered'].isin([True]), 1,0).astype(np.int64) df['Mature'] = np.where(df['Mature'].isin([True]), 1,0).astype(np.int64) # summarize DataFrame df.info()

# drop non-numeric columns df_numeric = df.drop(['Channel', 'Language'], axis=1) # retrieve correlation coefficients corrCoeff = np.corrcoef(df_numeric, rowvar=False) plt.figure(figsize=(10,6)) # create correlation heat map sns.heatmap(corrCoeff, annot=True, cmap='coolwarm') plt.title("Correlation Heatmap", fontsize=15) plt.xticks(np.arange(9)+0.5, df_numeric.columns, rotation=45, ha='right') plt.yticks(np.arange(9)+0.5, df_numeric.columns, rotation=0) plt.show()

# create copy of data df_model = df[['Average viewers', 'Watch time', 'Stream time', 'Followers']].copy() # select predictors X = df_model.drop(['Average viewers'], axis=1) # add constant/intercept X = sm.add_constant(X) # create response variable y = df_model['Average viewers'] # apply statsmodels OLS function model1 = sm.OLS(y, X) # fit the model results = model1.fit() results.summary()

# calculate VIF vif_data = pd.DataFrame() vif_data["variable"] = X.columns vif_data["VIF"] = [variance_inflation_factor(X.values, i) for i in range(len(X.columns))] vif_data

# generate pair plot sns.pairplot(df_model)

# create new copy for data with logarithmic values df_log = df[['Average viewers', 'Watch time', 'Stream time', 'Followers']].copy() df_log[['log(Average viewers)', 'log(Watch time)', 'log(Followers)']] = df_log[['Average viewers', 'Watch time', 'Followers']].apply(np.log) # select predictors X_log = df_log[['log(Watch time)', 'Stream time', 'log(Followers)']] # add constant/intercept X_log = sm.add_constant(X_log) # create response variable y_log = df_log['log(Average viewers)'] # apply statsmodels OLS function model2 = sm.OLS(y_log, X_log) # fit the model results = model2.fit() results.summary()

# split training and testing data X_train, X_test, y_train, y_test = train_test_split(X_new, y_new, test_size=0.3, random_state=26) # force coefficients to be positive lr = LinearRegression(positive=True) # fit training variables lr.fit(X_train, y_train) # generate predictions y_pred = lr.predict(X_test) # create DataFrame of actual and predicted number of average viewers df_model = pd.DataFrame({'Actual': y_test, 'Predicted': y_pred}) # calculate exponential of actual and predicted values np.exp(df_model.head(10))

# variables to calculate adjusted r-squared R2 = r2_score(y_test, y_pred) n = 0.3*1000 k = 3 print('Adjusted R-squared: ', (1 - ((1-R2) * (n-1) / (n-k-1)))) # calculate RMSE print('RMSE : ', np.sqrt(mean_squared_error(y_test,y_pred)))