Time Series Forecasting as Supervised Learning

Time series forecasting involves predicting future values based on previously observed data points. By reframing it as a supervised learning problem, you can leverage a variety of machine learning algorithms, both linear and nonlinear, to improve the forecasting accuracy. In this article, we will see how we can consider a supervised learning model.

What is Supervised Learning?

Supervised learning is a type of machine learning where an algorithm learns from labelled training data. It maps input variables (X) to output variables (y) using a function ?.

?=?(?)

The objective is to approximate this function so accurately that when new input data (X) is given, the algorithm can predict the corresponding output variable (y).

How to transform Time Series Data for Supervised Learning?

To apply supervised learning to time series forecasting, we need to transform the time series data into a suitable format. This involves creating input-output pairs from the sequential data using the sliding window technique.

The sliding window method involves using a fixed-size window of previous time steps to predict the next time step. This method is also known as lagged data.

For Example: Consider a time series with the following values:

Using a window size of 1, we can transform this time series into a supervised learning format:

Transformed for Supervised Learning:

In this format:

• ? represents the input features (previous time step's value).
• ? represents the target value (next time step's value).

This transformation allows us to apply supervised learning models to predict future values based on past observations. Here's a step-by-step breakdown of the process:

1. Sliding Window Method: For each time step, use the value of the previous time step as the input (X) and the current time step as the output (y).
2. Dataset Preparation: Create pairs of inputs and outputs from the time series data using the sliding window technique.

Implementing Supervised Learning Models on Time Series Data

Once the time series data is transformed, various supervised learning models can be applied, such as linear regression, decision trees, support vector machines (SVM), and neural networks.

1. Importing Necessary Libraries And Generating data

We import the necessary libraries and generate our data.

import numpy as np
from sklearn.linear_model import LinearRegression
from sklearn.model_selection import train_test_split

# Sample time series data
time_series = np.array([100, 110, 108, 115, 120])

# Prepare the dataset
window_size = 1
X, y = [], []
for i in range(len(time_series) - window_size):
    X.append(time_series[i:i + window_size])
    y.append(time_series[i + window_size])
X, y = np.array(X), np.array(y)

2. Splitting the data and training the model

We split the data into training and testing data, and train the linear regression model.

# Split into training and testing sets
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

# Train the model
model = LinearRegression()
model.fit(X_train, y_train)

3. Make Predictions on the Data

Leveraging our model, we make predictions.

# Make predictions
predictions = model.predict(X_test)
print("Predictions:", predictions)

Output:

Predictions: [116.55325444]

Conclusion

By framing this time series forecasting as a supervised learning problem, we can utilize a range of machine learning models to improve predictive accuracy. The sliding window method is a key technique in this transformation, enabling the creation of input-output pairs from sequential data. This approach is versatile and can be applied to both univariate and multivariate time series data, as well as for multi-step forecasting.