In [3]:
import pandas as pd
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt
%matplotlib inline
In [4]:
train = pd.read_csv('Competition_data.csv')
In [5]:
###
train.head()
Out[5]:
| Date | Close | |
|---|---|---|
| 0 | 1/1/2023 23:58:00 | 448.08 |
| 1 | 1/2/2023 23:58:00 | 448.08 |
| 2 | 1/3/2023 23:58:00 | 448.55 |
| 3 | 1/4/2023 23:58:00 | 449.01 |
| 4 | 1/5/2023 23:58:00 | 449.53 |
In [6]:
### The shape of the data
train.shape
Out[6]:
(486, 2)
In [7]:
train.info()
<class 'pandas.core.frame.DataFrame'> RangeIndex: 486 entries, 0 to 485 Data columns (total 2 columns): # Column Non-Null Count Dtype --- ------ -------------- ----- 0 Date 486 non-null object 1 Close 486 non-null float64 dtypes: float64(1), object(1) memory usage: 7.7+ KB
In [8]:
### Remove time
train['Date'] = train['Date'].apply(lambda x:x.split(' ')[0])
In [9]:
### Convert
train['Date'] = pd.to_datetime(train['Date'])
In [10]:
train
Out[10]:
| Date | Close | |
|---|---|---|
| 0 | 2023-01-01 | 448.08 |
| 1 | 2023-01-02 | 448.08 |
| 2 | 2023-01-03 | 448.55 |
| 3 | 2023-01-04 | 449.01 |
| 4 | 2023-01-05 | 449.53 |
| ... | ... | ... |
| 481 | 2024-05-02 | 1372.95 |
| 482 | 2024-05-03 | 1382.50 |
| 483 | 2024-05-04 | 1382.50 |
| 484 | 2024-05-05 | 1393.74 |
| 485 | 2024-05-06 | 1384.00 |
486 rows × 2 columns
In [11]:
train.set_index('Date').plot(title='Exchange Rate Trend');
plt.ylabel('Price');
Recently, the distribution of exchange prices has not been stable.
Create our test set¶
In [12]:
### The test set ranges from May 22 to June 4th
date_range = pd.date_range(start='2024-05-22', end='2024-06-04', freq='D')
In [13]:
test = pd.DataFrame(date_range, columns=['Date'])
In [14]:
test
Out[14]:
| Date | |
|---|---|
| 0 | 2024-05-22 |
| 1 | 2024-05-23 |
| 2 | 2024-05-24 |
| 3 | 2024-05-25 |
| 4 | 2024-05-26 |
| 5 | 2024-05-27 |
| 6 | 2024-05-28 |
| 7 | 2024-05-29 |
| 8 | 2024-05-30 |
| 9 | 2024-05-31 |
| 10 | 2024-06-01 |
| 11 | 2024-06-02 |
| 12 | 2024-06-03 |
| 13 | 2024-06-04 |
The test set ranges from May 22 to June 4th 2024.
Forecasting model¶
ARIMA models are a popular tool for time series forecasting, and can be implemented in Python using the statsmodels library.
In [15]:
from statsmodels.tsa.arima.model import ARIMA
from sklearn.metrics import mean_squared_error
In [16]:
### Cross validation
train_data = train[:480] ## we use the date from January 1 2023 to April 30 2024 to train our model
valid_data = train[480:] ## we use the date May 1 to May 6 to validate the performance of our model
In [17]:
train_close = train_data['Close'] ## Our target feature (Close)
valid_close = valid_data['Close']
In [18]:
historical_close = [x for x in train_close]
In [19]:
model = ARIMA(historical_close, order=(1,1,1))
In [20]:
model_fit = model.fit()
c:\Users\LENOVO\Music\Ad-Engagement-Forecasting\venv\lib\site-packages\statsmodels\tsa\statespace\sarimax.py:966: UserWarning: Non-stationary starting autoregressive parameters found. Using zeros as starting parameters.
warn('Non-stationary starting autoregressive parameters'
c:\Users\LENOVO\Music\Ad-Engagement-Forecasting\venv\lib\site-packages\statsmodels\tsa\statespace\sarimax.py:978: UserWarning: Non-invertible starting MA parameters found. Using zeros as starting parameters.
warn('Non-invertible starting MA parameters found.'
In [21]:
print(model_fit.summary())
SARIMAX Results
==============================================================================
Dep. Variable: y No. Observations: 480
Model: ARIMA(1, 1, 1) Log Likelihood -2266.694
Date: Sat, 25 May 2024 AIC 4539.388
Time: 11:46:44 BIC 4551.903
Sample: 0 HQIC 4544.308
- 480
Covariance Type: opg
==============================================================================
coef std err z P>|z| [0.025 0.975]
------------------------------------------------------------------------------
ar.L1 -0.9643 0.014 -66.538 0.000 -0.993 -0.936
ma.L1 0.9997 0.114 8.790 0.000 0.777 1.223
sigma2 751.2418 93.678 8.019 0.000 567.636 934.848
===================================================================================
Ljung-Box (L1) (Q): 0.01 Jarque-Bera (JB): 32721.07
Prob(Q): 0.90 Prob(JB): 0.00
Heteroskedasticity (H): 650.97 Skew: 4.28
Prob(H) (two-sided): 0.00 Kurtosis: 42.57
===================================================================================
Warnings:
[1] Covariance matrix calculated using the outer product of gradients (complex-step).
In [22]:
y_pred = model_fit.predict(start=len(train_data), end=len(train)-1)
In [23]:
mean_squared_error(valid_close, y_pred.round(2), squared=False)
Out[23]:
8.882015537027629
Validation score is 8.88
Test prediction¶
In [24]:
test['Date'].iloc[0]
Out[24]:
Timestamp('2024-05-22 00:00:00')
In [25]:
start_date = test['Date'].iloc[0]
end_date = test['Date'].iloc[-1]
prediction_dates = pd.date_range(start=start_date, end=end_date, freq='D')
In [26]:
periods = len(prediction_dates)
# Forecast
forecast = model_fit.get_forecast(steps=periods)
In [27]:
predicted_mean = forecast.predicted_mean
In [28]:
predicted_mean
Out[28]:
array([1389.26842887, 1389.97389697, 1389.29360052, 1389.94962346,
1389.31700793, 1389.92705124, 1389.33877475, 1389.90606108,
1389.35901597, 1389.88654208, 1389.37783852, 1389.86839114,
1389.39534182, 1389.85151237])
Make Submission (Make sure you round to 2 demical places)¶
In [29]:
submission = test
In [30]:
submission['Close'] = predicted_mean.round(2)
In [31]:
submission
Out[31]:
| Date | Close | |
|---|---|---|
| 0 | 2024-05-22 | 1389.27 |
| 1 | 2024-05-23 | 1389.97 |
| 2 | 2024-05-24 | 1389.29 |
| 3 | 2024-05-25 | 1389.95 |
| 4 | 2024-05-26 | 1389.32 |
| 5 | 2024-05-27 | 1389.93 |
| 6 | 2024-05-28 | 1389.34 |
| 7 | 2024-05-29 | 1389.91 |
| 8 | 2024-05-30 | 1389.36 |
| 9 | 2024-05-31 | 1389.89 |
| 10 | 2024-06-01 | 1389.38 |
| 11 | 2024-06-02 | 1389.87 |
| 12 | 2024-06-03 | 1389.40 |
| 13 | 2024-06-04 | 1389.85 |
In [32]:
## Submit to Zindi
submission.to_csv('base1.csv',index=False)
Recommendation¶
Data cleaning (if any).
Find the optimal parameters for the ARIMA model.
Try more sophisticated models like SARIMAX and FProphet.
Try the machine learning approach.
Ensemble Method.
Best of Luck in your Forecast¶
In [ ]: