电子游戏销售数据可视化分析及预测

对全球电子游戏销售数据进行可视化展示，并使用支持向量回归进行预测分析

数据可视化部分

In [1]:

import numpy as np 
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns

In [20]:

df = pd.read_csv("/home/mw/input/games6014/games_utf8.csv")

In [9]:

df.head(5)

Out[9]:

	index	Name	Platform	Year_of_Release	Genre	Publisher	NA_Sales	EU_Sales	JP_Sales	Other_Sales	Global_Sales	Critic_Score	Critic_Count	User_Score	User_Count	Developer	Rating
0	0	Wii Sports	Wii	2006	Sports	Nintendo	41.36	28.96	3.77	8.45	82.53	76.0	51.0	8	322.0	Nintendo	E
1	1	Super Mario Bros.	NES	1985	Platform	Nintendo	29.08	3.58	6.81	0.77	40.24	NaN	NaN	NaN	NaN	NaN	NaN
2	2	Mario Kart Wii	Wii	2008	Racing	Nintendo	15.68	12.76	3.79	3.29	35.52	82.0	73.0	8.3	709.0	Nintendo	E
3	3	Wii Sports Resort	Wii	2009	Sports	Nintendo	15.61	10.93	3.28	2.95	32.77	80.0	73.0	8	192.0	Nintendo	E
4	4	Pokemon Red/Pokemon Blue	GB	1996	Role-Playing	Nintendo	11.27	8.89	10.22	1.00	31.37	NaN	NaN	NaN	NaN	NaN	NaN

In [10]:

df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 16928 entries, 0 to 16927
Data columns (total 17 columns):
index              16928 non-null int64
Name               16926 non-null object
Platform           16928 non-null object
Year_of_Release    16655 non-null object
Genre              16926 non-null object
Publisher          16873 non-null object
NA_Sales           16928 non-null float64
EU_Sales           16928 non-null float64
JP_Sales           16928 non-null float64
Other_Sales        16928 non-null float64
Global_Sales       16926 non-null float64
Critic_Score       8260 non-null float64
Critic_Count       8260 non-null float64
User_Score         10159 non-null object
User_Count         7718 non-null float64
Developer          10240 non-null object
Rating             10092 non-null object
dtypes: float64(8), int64(1), object(8)
memory usage: 2.2+ MB

查看各个字段不同数量

可以看到有
有16928条销售信息
有11562个不同的游戏
有33个不同的游戏平台

In [11]:

df.nunique()

Out[11]:

index              16928
Name               11562
Platform              33
Year_of_Release       40
Genre                 14
Publisher            582
NA_Sales             402
EU_Sales             307
JP_Sales             244
Other_Sales          155
Global_Sales         629
Critic_Score          82
Critic_Count         106
User_Score            96
User_Count           888
Developer           1696
Rating                 8
dtype: int64

最畅销的游戏

在将不同平台上的相同游戏分组后，以下是前 5 名最畅销的游戏及其以百万计的销售数量。

In [13]:

df.groupby('Name')['Global_Sales'].sum().sort_values(ascending=False).head()

Out[13]:

Name
Wii Sports            82.53
Grand Theft Auto V    56.57
Super Mario Bros.     45.31
Tetris                35.84
Mario Kart Wii        35.52
Name: Global_Sales, dtype: float64

In [14]:

df.groupby('Name')['Global_Sales'].sum().describe().T

Out[14]:

count    11562.000000
mean         0.789696
std          2.284268
min          0.000000
25%          0.060000
50%          0.190000
75%          0.610000
max         82.530000
Name: Global_Sales, dtype: float64

数据清洗

保留需要的字段并去空值

In [21]:

df = df.iloc[:,0:11] 
df = df.dropna().sort_values(by='Year_of_Release')

绘制描述游戏销售变化的折线图。

先对数据进行处理分析方便可视化

In [26]:

df_year = df.groupby('Year_of_Release').sum()
df_year['count'] = df.groupby('Year_of_Release').count()['index']
df_year = df_year.reset_index()
df_year['sales_to_count'] = df_year['Global_Sales']/df_year['count']
df_year['Year_of_Release'] = df_year['Year_of_Release'].astype('int32')
df_year

Out[26]:

	Year_of_Release	index	NA_Sales	EU_Sales	JP_Sales	Other_Sales	Global_Sales	count	sales_to_count
0	1980	29746	10.59	0.67	0.00	0.12	11.38	9	1.264444
1	1981	189885	33.40	1.96	0.00	0.32	35.77	46	0.777609
2	1982	199233	28.66	1.75	0.00	0.34	30.73	39	0.787949
3	1983	56570	7.76	0.80	8.10	0.14	16.79	17	0.987647
4	1984	22869	33.28	2.10	14.27	0.70	50.36	14	3.597143
5	1985	55624	33.73	4.74	14.56	0.92	53.94	14	3.852857
6	1986	35906	12.50	2.84	19.81	1.93	37.07	21	1.765238
7	1987	54624	8.46	1.41	11.63	0.20	21.74	16	1.358750
8	1988	37250	23.87	6.59	15.76	0.99	47.22	15	3.148000
9	1989	40131	45.15	8.44	18.36	1.50	73.45	17	4.320588
10	1990	25062	25.46	7.63	14.88	1.40	49.39	16	3.086875
11	1991	229258	13.08	4.00	15.03	0.75	32.86	42	0.782381
12	1992	208574	33.89	11.71	28.91	1.65	76.17	43	1.771395
13	1993	395706	15.42	4.65	25.65	0.89	46.60	61	0.763934
14	1994	868973	28.16	14.88	33.99	2.20	79.18	121	0.654380
15	1995	1931422	24.84	14.90	46.99	2.64	89.36	222	0.402523
16	1996	1979410	90.34	49.85	59.62	8.03	207.83	267	0.778390
17	1997	1964191	95.79	49.03	49.46	9.29	203.47	293	0.694437
18	1998	2742972	137.91	71.32	52.86	11.64	273.86	386	0.709482
19	1999	2271203	127.94	63.50	52.57	10.18	254.34	343	0.741516
20	2000	2458768	96.14	53.51	42.77	11.75	204.10	354	0.576554
21	2001	3523887	180.42	96.74	41.92	23.15	342.25	490	0.698469
22	2002	7040098	222.19	114.10	42.60	28.52	407.95	841	0.485077
23	2003	6270393	200.59	106.57	35.82	26.42	369.66	784	0.471505
24	2004	5853176	225.21	110.89	41.73	48.68	426.83	754	0.566088
25	2005	7836046	245.33	122.02	54.49	40.71	463.04	944	0.490508
26	2006	9712227	267.34	131.03	75.11	55.15	529.11	1014	0.521805
27	2007	10522058	311.97	159.01	61.37	77.36	610.33	1207	0.505659
28	2008	12393789	356.60	184.03	60.67	83.17	684.76	1441	0.475198
29	2009	12809114	345.41	194.51	62.33	76.12	678.44	1448	0.468536
30	2010	11308061	305.09	173.47	60.66	59.32	599.05	1266	0.473183
31	2011	10270400	244.76	167.68	55.12	54.42	522.17	1146	0.455646
32	2012	5714648	159.17	117.52	53.26	37.05	367.00	662	0.554381
33	2013	4537392	154.62	122.69	47.69	38.72	363.72	549	0.662514
34	2014	5175494	137.61	127.50	41.08	38.08	344.23	595	0.578538
35	2015	6019607	110.64	100.81	34.45	31.75	277.71	612	0.453775
36	2016	5670596	48.21	52.96	22.17	14.98	138.48	505	0.274218
37	2017	46693	0.00	0.00	0.06	0.00	0.06	3	0.020000
38	2020	5936	0.27	0.00	0.00	0.02	0.29	1	0.290000

在左侧，显示全球总销量，即当年所有游戏的全球销量之和。
在右侧，显示每款游戏的全球销售额或每款游戏的平均销售额。

In [32]:

fig, (ax0,ax1) = plt.subplots(nrows=1,ncols=2,figsize=(15,5))
sns.lineplot(df_year['Year_of_Release'], df_year['Global_Sales'],ax=ax0)
sns.lineplot(df_year['Year_of_Release'], df_year['sales_to_count'],ax=ax1)
plt.show()

绘制游戏类型与销售额的比较图

In [27]:

df_genre = df.groupby('Genre').sum()
df_genre['count'] = df.groupby('Genre').count()['index']
df_genre = df_genre.reset_index()
df_genre['genre_to_count'] = df_genre['Global_Sales']/df_genre['count']
df_genre

Out[27]:

	Genre	index	NA_Sales	EU_Sales	JP_Sales	Other_Sales	Global_Sales	count	genre_to_count
0	Action	27483058	877.67	517.89	162.21	184.62	1743.46	3346	0.521058
1	Adventure	15170655	102.19	63.73	53.02	16.60	235.65	1299	0.181409
2	Fighting	6635454	231.01	103.54	91.63	36.93	463.06	851	0.544136
3	Misc	15076819	408.60	216.54	108.27	76.14	810.20	1720	0.471047
4	Platform	6335987	455.94	206.14	133.71	52.38	848.51	894	0.949116
5	Puzzle	5558196	121.14	49.78	57.30	12.30	240.96	571	0.421996
6	Racing	9985753	370.47	244.40	59.10	77.92	752.08	1240	0.606516
7	Role-Playing	12300382	338.04	194.21	360.91	61.59	954.49	1501	0.635903
8	Shooter	9912089	600.66	326.38	40.56	107.48	1075.56	1320	0.814818
9	Simulation	7489053	183.35	114.01	63.64	30.78	391.92	861	0.455192
10	Sports	17690030	683.73	375.80	135.20	133.54	1329.00	2338	0.568435
11	Strategy	6869516	69.00	45.39	50.20	10.92	175.80	677	0.259675

In [33]:

fig, (ax0,ax1) = plt.subplots(nrows=1,ncols=2,figsize=(15,5))
sns.barplot(df_genre['Genre'], df_genre['Global_Sales'],ax=ax0)
ax0.set_xticklabels(ax0.get_xticklabels(), rotation=45)
sns.barplot(df_genre['Genre'], df_genre['genre_to_count'],ax=ax1)
ax1.set_xticklabels(ax1.get_xticklabels(), rotation=45)
plt.show()

销售预测

编码

get_dummies 是利用pandas实现one hot encode的方式

In [36]:

df.head(1)

Out[36]:

	index	Name	Platform	Year_of_Release	Genre	Publisher	NA_Sales	EU_Sales	JP_Sales	Other_Sales	Global_Sales
5360	5360	Freeway	2600	1980	Action	Activision	0.32	0.02	0.0	0.0	0.34

In [38]:

df_code = pd.get_dummies(df.iloc[:,2:11], drop_first=True)
df_code.head(1)

Out[38]:

	NA_Sales	EU_Sales	JP_Sales	Other_Sales	Global_Sales	Platform_3DO	Platform_3DS	Platform_DC	Platform_DS	Platform_GB	...	Publisher_Zushi Games	Publisher_bitComposer Games	Publisher_dramatic create	Publisher_fonfun	Publisher_iWin	Publisher_id Software	Publisher_imageepoch Inc.	Publisher_inXile Entertainment	Publisher_mixi, Inc	Publisher_responDESIGN
5360	0.32	0.02	0.0	0.0	0.34	0	0	0	0	0	...	0	0	0	0	0	0	0	0	0	0

1 rows × 662 columns

训练集测试集划分

划分训练集和测试集并标准化数据

In [39]:

from sklearn.model_selection import train_test_split, cross_val_score, KFold
from sklearn.preprocessing import StandardScaler

X = df_code.drop('Global_Sales', axis=1).values
y = df_code['Global_Sales'].values

# 切分训练集和测试集
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.2, random_state = 43)

In [40]:

# 数据标准化
scaler = StandardScaler()
X_train_scaled = scaler.fit_transform(X_train)
X_test_scaled = scaler.transform(X_test)

In [41]:

X_train_scaled

Out[41]:

array([[ 0.58116251, -0.11136746, -0.26848048, ..., -0.00867338,
        -0.00867338, -0.01226648],
       [-0.16340037, -0.09217906, -0.26848048, ..., -0.00867338,
        -0.00867338, -0.01226648],
       [-0.2106742 , -0.16893264, -0.26848048, ..., -0.00867338,
        -0.00867338, -0.01226648],
       ...,
       [-0.31704033, -0.28406301, -0.09774392, ..., -0.00867338,
        -0.00867338, -0.01226648],
       [-0.31704033, -0.28406301, -0.23433317, ..., -0.00867338,
        -0.00867338, -0.01226648],
       [ 0.13206109,  0.48347278, -0.23433317, ..., -0.00867338,
        -0.00867338, -0.01226648]])

In [42]:

X_test_scaled

Out[42]:

array([[-0.31704033, -0.28406301,  0.82423352, ..., -0.00867338,
        -0.00867338, -0.01226648],
       [-0.31704033, -0.20730943, -0.26848048, ..., -0.00867338,
        -0.00867338, -0.01226648],
       [-0.00976041,  0.27240044, -0.20018586, ..., -0.00867338,
        -0.00867338, -0.01226648],
       ...,
       [-0.31704033, -0.16893264, -0.26848048, ..., -0.00867338,
        -0.00867338, -0.01226648],
       [-0.31704033, -0.28406301, -0.16603855, ..., -0.00867338,
        -0.00867338, -0.01226648],
       [-0.04521578, -0.16893264, -0.26848048, ..., -0.00867338,
        -0.00867338, -0.01226648]])

SVR模型预测

创建一个SVR模型并使用测试集数据进行预测，最后输出预测的均方误差

In [44]:

from sklearn.svm import SVR
from sklearn.metrics import mean_squared_error

# 创建SVR模型
svr_model = SVR()

# 使用X_train_scaled和y_train来训练模型
svr_model.fit(X_train_scaled, y_train)

# 使用训练好的模型对X_test_scaled进行预测
y_pred = svr_model.predict(X_test_scaled)

# 计算预测均方误差
mse = mean_squared_error(y_test, y_pred)
print("均方误差：", mse)

/opt/conda/lib/python3.6/site-packages/sklearn/svm/base.py:193: FutureWarning: The default value of gamma will change from 'auto' to 'scale' in version 0.22 to account better for unscaled features. Set gamma explicitly to 'auto' or 'scale' to avoid this warning.
  "avoid this warning.", FutureWarning)

均方误差： 0.44936424854549994

结果可视化

黑色点代表真实值，蓝色线代表SVR的预测结果

In [45]:

import matplotlib.pyplot as plt

# 绘制真实值
plt.scatter(range(len(y_test)), y_test, color='black', label='True values')

# 绘制预测值
plt.plot(range(len(y_test)), y_pred, color='blue', linewidth=3, label='Predicted values')

plt.xlabel('Sample')
plt.ylabel('Value')
plt.title('SVR Prediction Results')
plt.legend()
plt.show()