Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Applied Machine Learning Methods with Long-Short Term Memory Based Recurrent Neural Networks for Multivariate Temperature Prediction

Bojan Lukić

TL;DR

The paper tackles multivariate weather time-series prediction by constructing a deep LSTM-based recurrent neural network trained in Python with TensorFlow/Keras on historical weather data (temp, hum, airpr, solrad, windvel, winddir). It systematically evaluates multi-step forecasts at 7-day, 1-day, and 12-hour horizons, exploring data preparation, feature selection, and regularization to mitigate overfitting. Key findings show that removing volatile features (winddir) and applying $L2$ regularization substantially improve forecasting accuracy, with final metrics around $\text{RMSE}=0.26$ and $\text{MAE}=0.26$ for short horizons and sub-0.5°C max deviations for 12-hour forecasts. Despite improvements, the work highlights persistent challenges for longer-range predictions and underlines data quality and representativeness as critical factors, offering a practical, open-source workflow for deploying LSTM-based time-series predictors in real-world weather forecasting. The results demonstrate that with careful data curation and hyperparameter tuning, deep LSTM models can provide useful short-term forecasts in multivariate settings.

Abstract

This paper gives an overview on how to develop a dense and deep neural network for making a time series prediction. First, the history and cornerstones in Artificial Intelligence and Machine Learning will be presented. After a short introduction to the theory of Artificial Intelligence and Machine Learning, the paper will go deeper into the techniques for conducting a time series prediction with different models of neural networks. For this project, Python's development environment Jupyter, extended with the TensorFlow package and deep-learning application Keras is used. The system setup and project framework are explained in more detail before discussing the time series prediction. The main part shows an applied example of time series prediction with weather data. For this work, a deep recurrent neural network with Long Short-Term Memory cells is used to conduct the time series prediction. The results and evaluation of the work show that a weather prediction with deep neural networks can be successful for a short time period. However, there are some drawbacks and limitations with time series prediction, which will be discussed towards the end of the paper.

Applied Machine Learning Methods with Long-Short Term Memory Based Recurrent Neural Networks for Multivariate Temperature Prediction

TL;DR

The paper tackles multivariate weather time-series prediction by constructing a deep LSTM-based recurrent neural network trained in Python with TensorFlow/Keras on historical weather data (temp, hum, airpr, solrad, windvel, winddir). It systematically evaluates multi-step forecasts at 7-day, 1-day, and 12-hour horizons, exploring data preparation, feature selection, and regularization to mitigate overfitting. Key findings show that removing volatile features (winddir) and applying regularization substantially improve forecasting accuracy, with final metrics around and for short horizons and sub-0.5°C max deviations for 12-hour forecasts. Despite improvements, the work highlights persistent challenges for longer-range predictions and underlines data quality and representativeness as critical factors, offering a practical, open-source workflow for deploying LSTM-based time-series predictors in real-world weather forecasting. The results demonstrate that with careful data curation and hyperparameter tuning, deep LSTM models can provide useful short-term forecasts in multivariate settings.

Abstract

This paper gives an overview on how to develop a dense and deep neural network for making a time series prediction. First, the history and cornerstones in Artificial Intelligence and Machine Learning will be presented. After a short introduction to the theory of Artificial Intelligence and Machine Learning, the paper will go deeper into the techniques for conducting a time series prediction with different models of neural networks. For this project, Python's development environment Jupyter, extended with the TensorFlow package and deep-learning application Keras is used. The system setup and project framework are explained in more detail before discussing the time series prediction. The main part shows an applied example of time series prediction with weather data. For this work, a deep recurrent neural network with Long Short-Term Memory cells is used to conduct the time series prediction. The results and evaluation of the work show that a weather prediction with deep neural networks can be successful for a short time period. However, there are some drawbacks and limitations with time series prediction, which will be discussed towards the end of the paper.

Paper Structure

This paper contains 21 sections, 2 equations, 16 figures.

Table of Contents

  1. Introduction
  2. Related Work
  3. Background of Machine Learning and Artificial Intelligence
  4. Machine Learning
  5. Artificial Intelligence and Neural Networks
  6. Problem statement and system setup
  7. Different approaches to time series prediction
  8. Recurrent Neurons and Layers with memory cells
  9. Deep Recurrent Neural Networks
  10. Long Short-Term Memory
  11. Conducting a time series prediction with the LSTM model
  12. Setup
  13. Forecasting a 7-day period
  14. Forecasting a 1-day period
  15. Forecasting a 12-hour period
  16. ...and 6 more sections

Figures (16)

  • Figure 1: The Machine Learning approach [2].
  • Figure 2: ANNs performing simple logical computations [2].
  • Figure 3: Threshold logic unit: an artificial neuron which computes a weighted sum of its inputs then applies a step function [2].
  • Figure 4: An MLP that solves the XOR classification problem [2].
  • Figure 5: A recurrent neuron (left) unrolled through time (right) [2].
  • ...and 11 more figures