seasonal_decompose

[Notice] [Montly_Data]

Monthly Data

• Australian anti-diabetic monthly sales data will be used

https://raw.githubusercontent.com/selva86/datasets/master/a10.csv

• Used for performance data such as monthly sales, subscribers, etc. in all companies

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import warnings
warnings.filterwarnings("ignore")

from statsmodels.tsa.seasonal import seasonal_decompose

• Import data. plottig.

• Separation of seasonal factors. Create a table containing Trend, Seasonl, and Residual.

• Insight Derivation

  1. Average monthly growth rate

  2. Seasonal Factor Analysis

  3. Whether residual increases or decreases

---

# convert date type to date 
df=pd.read_csv('https://raw.githubusercontent.com/selva86/datasets/master/a10.csv', parse_dates=['date'], index_col='date')
df.head()

	value
date
1991-07-01	3.526591
1991-08-01	3.180891
1991-09-01	3.252221
1991-10-01	3.611003
1991-11-01	3.565869

df.value.plot()

<AxesSubplot:xlabel='date'>

df=df.loc[df.index>'1999-12-31']

# Set tow_sided to false since recent data is more important
result= seasonal_decompose(df, model='additive', two_sided=False)

result.plot()
plt.show()

df_re=pd.concat([result.observed, result.trend, result.seasonal, result.resid], axis=1)

df_re.head()

	0	trend	seasonal	resid
date
2000-01-01	12.511462	NaN	5.061367	NaN
2000-02-01	7.457199	NaN	-3.307973	NaN
2000-03-01	8.591191	NaN	-2.293969	NaN
2000-04-01	8.474000	NaN	-2.264396	NaN
2000-05-01	9.386803	NaN	-0.711368	NaN

df_re.columns=['obs', 'trend', 'seasonal', 'resid']
df_re.dropna(inplace=True)

df_re.head()

	obs	trend	seasonal	resid
date
2001-01-01	14.497581	10.290804	5.061367	-0.854590
2001-02-01	8.049275	10.398229	-3.307973	0.959019
2001-03-01	10.312891	10.494636	-2.293969	2.112225
2001-04-01	9.753358	10.619680	-2.264396	1.398075
2001-05-01	10.850382	10.733969	-0.711368	0.827781

df_re.head(24)
df_re['year']=df_re.index.year

df_re.head()

	obs	trend	seasonal	resid	year
date
2001-01-01	14.497581	10.290804	5.061367	-0.854590	2001
2001-02-01	8.049275	10.398229	-3.307973	0.959019	2001
2001-03-01	10.312891	10.494636	-2.293969	2.112225	2001
2001-04-01	9.753358	10.619680	-2.264396	1.398075	2001
2001-05-01	10.850382	10.733969	-0.711368	0.827781	2001

plt.figure(figsize=(16,6))
plt.plot(df_re.obs)
plt.plot(df_re.trend)
plt.plot(df_re.seasonal+df_re.trend)
plt.legend(['observed', 'trend', 'seasonal+trend'])
plt.tight_layout()
plt.show()
# In 2006 and 2007, there is a tendency to deviate from the ordinary cycle. (residual is large.)

# how to see percentage of trend
df_re['trend'].pct_change().dropna().plot(kind = 'bar', figsize = (16, 5))

<AxesSubplot:xlabel='date'>