회귀 분석을 이용한 앨범 판매량 분석 및 예측

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
from scipy import stats
import statsmodels.api as sm
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_squared_error
plt.style.use('fivethirtyeight')

sales = pd.read_csv('Album_sales_2.txt',sep='\t')
sales

1

sales.info()

1

sales.describe()

1
2
3
4
5

plt.style.use('seaborn-pastel')
plt.figure(figsize=(10,10))
for i,j in enumerate(sales.columns):
    plt.subplot(2,2,i+1)
    sns.histplot(sales[j])

1

sns.pairplot(sales)

1

sns.heatmap(sales.corr(),annot=True,cmap='Blues')

1

x_train, x_test, y_train, y_test = train_test_split(sales.iloc[:,[0,2,3]], sales['sales'],)

1

x_data = sm.add_constant(x_train, has_constant = "add")

1
2
3
4

# const넣음
model = sm.OLS(y_train,x_data)
result = model.fit()
result.summary()

1
2

y_pred = result.predict(ols.add_constant(x_test, has_constant = "add"))
mean_squared_error(y_pred,y_test)

1
2
3
4

# 안 넣음
model2 = sm.OLS(y_train,x_train)
result2 = model2.fit()
result2.summary()

1
2

y_pred = result2.predict(x_test)
mean_squared_error(y_pred,y_test)

1
2

plt.plot(y_test.values)
plt.plot(y_pred.values)

1
2

from statsmodels.stats.outliers_influence import variance_inflation_factor 
[variance_inflation_factor(x_train.values,i) for i in range(3)]

1
2
3
4
5
6

from itertools import combinations
joint = []
for i in range(1,4):
    k = combinations(x_train.columns,i)
    for j in k:
        joint.append(j)

1
2
3
4
5
6

stepw = []
for i in joint:
    model = sm.OLS(y_train,x_train[list(i)])
    result = model.fit()
    y_pred = result.predict(x_test[list(i)])
    stepw.append([list(i),result.aic,mean_squared_error(y_pred,y_test)])

1
2
3

stepw_frame = pd.DataFrame(stepw, columns=['Variables','AIC','MSE'])
stepw_frame.set_index('Variables',inplace=True)
stepw_frame

1

result2.conf_int()

1
2

# 독립성
sns.regplot(result2.fittedvalues, result2.resid, lowess=True, line_kws={'color': 'red'})

1
2

#정규성
stats.probplot(result2.resid,plot=plt);

1

stats.shapiro(result2.resid)

1
2
3

infl = result2.get_influence()
sm_fr = infl.summary_frame()
sm_fr

1
2
3
4
5
6
7

album_sales_diagnostics = pd.concat([x_train,y_train],axis=1)

album_sales_diagnostics['cook'] = infl.cooks_distance[0]
album_sales_diagnostics['resid_std'] = infl.resid_studentized
album_sales_diagnostics[['dfbeta_adverts','dfbeta_airplay','dfbeta_attract']] = infl.dfbeta
album_sales_diagnostics['covariance'] = infl.cov_ratio
album_sales_diagnostics.head()

1
2

album.resid_std = np.abs(album.resid_std)
album.sort_values(by = ['cook','resid_std'],ascending=[False, False])