• Missing values in time series forecasting
  • Libraries
  • Data
  • Impute missing values
  • Using imputed values in model training
  • Give weight of zero to imputed values
  • Session information
  • Citation


More about forecasting in cienciadedatos.net


Missing values in time series forecasting

In many real use cases of forecasting, although historical data are available, it is common for the time series to be incomplete. The presence of missing values in the data is a major problem since most forecasting algorithms require the time series to be complete in order to train a model.

A commonly employed strategy to overcome this problem is to impute missing values before training the model, for example, using a moving average. However, the quality of the imputations may not be good, impairing the training of the model. One way to improve the imputation strategy is to combine it with weighted time series forecasting. The latter consists of reducing the weight of the imputed observations and thus their influence during model training.

This document shows two examples of how skforecast makes it easy to apply this strategy.

Libraries

# Data manipulation
# ==============================================================================
import numpy as np
import pandas as pd
from skforecast.datasets import fetch_dataset

# Plots
# ==============================================================================
import matplotlib.pyplot as plt
from skforecast.plot import set_dark_theme

# Modeling and Forecasting
# ==============================================================================
import sklearn
import skforecast
from lightgbm import LGBMRegressor
from skforecast.recursive import ForecasterRecursive
from skforecast.model_selection import TimeSeriesFold
from skforecast.model_selection import backtesting_forecaster

# Warnings configuration
# ==============================================================================
import warnings

color = '\033[1m\033[38;5;208m' 
print(f"{color}Version skforecast: {skforecast.__version__}")
print(f"{color}Version scikit-learn: {sklearn.__version__}")
print(f"{color}Version pandas: {pd.__version__}")
print(f"{color}Version numpy: {np.__version__}")
Version skforecast: 0.15.1
Version scikit-learn: 1.5.2
Version pandas: 2.2.3
Version numpy: 2.0.2

✎ Note

In this document, a forecaster of type ForecasterAutoreg is used. The same strategy can be applied with any forecaster from skforecast.

Data

# Data download
# ==============================================================================
data = fetch_dataset('bicimad')
data
bicimad
-------
This dataset contains the daily users of the bicycle rental service (BiciMad) in
the city of Madrid (Spain) from 2014-06-23 to 2022-09-30.
The original data was obtained from: Portal de datos abiertos del Ayuntamiento
de Madrid https://datos.madrid.es/portal/site/egob
Shape of the dataset: (3022, 1)
users
date
2014-06-23 99
2014-06-24 72
2014-06-25 119
2014-06-26 135
2014-06-27 149
... ...
2022-09-26 12340
2022-09-27 13888
2022-09-28 14239
2022-09-29 11574
2022-09-30 12957

3022 rows × 1 columns

# Generating gaps with missing values
# ==============================================================================
gaps = [
    ['2020-09-01', '2020-10-10'],
    ['2020-11-08', '2020-12-15'],
]

for gap in gaps:
    data.loc[gap[0]:gap[1]] = np.nan
# Split data into train-test
# ==============================================================================
data = data.loc['2020-06-01':'2021-06-01'].copy()
end_train = '2021-03-01'
data_train = data.loc[: end_train, :]
data_test  = data.loc[end_train:, :]

print(f"Dates train : {data_train.index.min()} --- {data_train.index.max()}  (n={len(data_train)})")
print(f"Dates test  : {data_test.index.min()} --- {data_test.index.max()}  (n={len(data_test)})")
Dates train : 2020-06-01 00:00:00 --- 2021-03-01 00:00:00  (n=274)
Dates test  : 2021-03-01 00:00:00 --- 2021-06-01 00:00:00  (n=93)
# Time series plot
# ==============================================================================
set_dark_theme()
fig, ax = plt.subplots(figsize=(7, 3))
data_train.users.plot(ax=ax, label='train', linewidth=1)
data_test.users.plot(ax=ax, label='test', linewidth=1)

for gap in gaps:
    ax.plot(
        [pd.to_datetime(gap[0]), pd.to_datetime(gap[1])],
        [data.users[pd.to_datetime(gap[0]) - pd.Timedelta(days=1)],
         data.users[pd.to_datetime(gap[1]) + pd.Timedelta(days=1)]],
        color = 'red',
        linestyle = '--',
        label = 'gap'
        )
ax.set_xlabel("")
ax.set_title('Number of users BiciMAD')
handles, labels = plt.gca().get_legend_handles_labels()
by_label = dict(zip(labels, handles))
ax.legend(by_label.values(), by_label.keys(), loc='lower right');

Impute missing values

# Value imputation using linear interpolation
# ======================================================================================
data['users_imputed'] = data['users'].interpolate(method='linear')
data_train = data.loc[: end_train, :]
data_test  = data.loc[end_train:, :]

Using imputed values in model training

# Create recursive multi-step forecaster (ForecasterRecursive)
# ==============================================================================
forecaster = ForecasterRecursive(
                 regressor = LGBMRegressor(random_state=123, verbose=-1),
                 lags      = 14
             )

# Backtesting: predict next 7 days at a time.
# ==============================================================================
cv = TimeSeriesFold(
        steps              = 7,
        initial_train_size = len(data.loc[:end_train]),
        refit              = True,
        fixed_train_size   = False
    )
metric, predictions = backtesting_forecaster(
                        forecaster = forecaster,
                        y          = data.users_imputed,
                        cv         = cv,
                        metric     = 'mean_absolute_error',
                      )
display(metric)
predictions.head(4)
  0%|          | 0/14 [00:00<?, ?it/s]
mean_absolute_error
0 2151.339364
pred
2021-03-02 9679.561409
2021-03-03 10556.841280
2021-03-04 8922.423792
2021-03-05 8874.277159

Give weight of zero to imputed values

To minimize the influence on the model of imputed values, a custom function is defined to create weights following the rules:

  • Weight of 0 if the index date has been imputed or is within 14 days ahead of an imputed day.

  • Weight of 1 otherwise.

If an observation has a weight of 0, it has no influence at all during model training.

✎ Note

Imputed values should neither participate in the training process as a target nor as a predictor (lag). Therefore, values within a window size as large as the lags used should also be excluded.
# Custom function to create weights
# ==============================================================================
def custom_weights(index):
    """
    Return 0 if index is in any gap.
    """
    gaps = [
        ['2020-09-01', '2020-10-10'],
        ['2020-11-08', '2020-12-15'],
    ]
    
    missing_dates = [pd.date_range(
                        start = pd.to_datetime(gap[0]) + pd.Timedelta('14d'),
                        end   = pd.to_datetime(gap[1]) + pd.Timedelta('14d'),
                        freq  = 'D'
                    ) for gap in gaps]
    missing_dates = pd.DatetimeIndex(np.concatenate(missing_dates))   
    weights = np.where(index.isin(missing_dates), 0, 1)

    return weights

Again, a ForecasterAutoreg is trained but this time including the custom_weights function.

# Create recursive multi-step forecaster (ForecasterRecursive)
# ==============================================================================
forecaster = ForecasterRecursive(
                 regressor   = LGBMRegressor(random_state=123, verbose=-1),
                 lags        = 14,
                 weight_func = custom_weights
             )

# Backtesting: predict next 7 days at a time.
# ==============================================================================
metric, predictions = backtesting_forecaster(
                            forecaster = forecaster,
                            y          = data.users_imputed,
                            cv         = cv,
                            metric     = 'mean_absolute_error',
                       )
display(metric)
predictions.head(4)
  0%|          | 0/14 [00:00<?, ?it/s]
mean_absolute_error
0 1904.830714
pred
2021-03-02 10524.159747
2021-03-03 10087.283682
2021-03-04 8882.926166
2021-03-05 9474.810215

Giving a weight of 0 to the imputed values (excluding it from the model training) improves the forecasting performance.

Session information

import session_info
session_info.show(html=False)
-----
lightgbm            4.6.0
matplotlib          3.10.1
numpy               2.0.2
pandas              2.2.3
session_info        1.0.0
skforecast          0.15.1
sklearn             1.5.2
-----
IPython             9.0.2
jupyter_client      8.6.3
jupyter_core        5.7.2
notebook            6.4.12
-----
Python 3.12.9 | packaged by Anaconda, Inc. | (main, Feb  6 2025, 18:56:27) [GCC 11.2.0]
Linux-5.15.0-1077-aws-x86_64-with-glibc2.31
-----
Session information updated at 2025-03-26 16:48

Citation

How to cite this document

If you use this document or any part of it, please acknowledge the source, thank you!

Forecasting time series with missing values by Joaquín Amat Rodrigo and Javier Escobar Ortiz, available under a Attribution-NonCommercial-ShareAlike 4.0 International at https://www.cienciadedatos.net/documentos/py46-forecasting-time-series-missing-values

How to cite skforecast

If you use skforecast for a publication, we would appreciate it if you cite the published software.

Zenodo:

Amat Rodrigo, Joaquin, & Escobar Ortiz, Javier. (2024). skforecast (v0.15.1). Zenodo. https://doi.org/10.5281/zenodo.8382788

APA:

Amat Rodrigo, J., & Escobar Ortiz, J. (2024). skforecast (Version 0.15.1) [Computer software]. https://doi.org/10.5281/zenodo.8382788

BibTeX:

@software{skforecast, author = {Amat Rodrigo, Joaquin and Escobar Ortiz, Javier}, title = {skforecast}, version = {0.15.1}, month = {03}, year = {2025}, license = {BSD-3-Clause}, url = {https://skforecast.org/}, doi = {10.5281/zenodo.8382788} }


Did you like the article? Your support is important

Your contribution will help me to continue generating free educational content. Many thanks! 😊

Become a GitHub Sponsor Become a GitHub Sponsor

Creative Commons Licence

This work by Joaquín Amat Rodrigo and Javier Escobar Ortiz is licensed under a Attribution-NonCommercial-ShareAlike 4.0 International.

Allowed:

  • Share: copy and redistribute the material in any medium or format.

  • Adapt: remix, transform, and build upon the material.

Under the following terms:

  • Attribution: You must give appropriate credit, provide a link to the license, and indicate if changes were made. You may do so in any reasonable manner, but not in any way that suggests the licensor endorses you or your use.

  • NonCommercial: You may not use the material for commercial purposes.

  • ShareAlike: If you remix, transform, or build upon the material, you must distribute your contributions under the same license as the original.