Krishnan et al. (2026) XAITempSpikeDetector: XAI-Powered Temperature Prediction and Spike Detection, A Data-Driven Approach

Identification

• Journal: Lecture notes in networks and systems
• Year: 2026
• Date: 2026-01-01
• Authors: Archana Krishnan, K. A. Rafidha Rehiman, M. K. Sabu
• DOI: 10.1007/978-3-032-06250-5_28

Research Groups

• Department of Computer Applications, Cochin University of Science and Technology, Kochi, Kerala, India

Short Summary

This paper proposes XAI-TempSpikeDetector, a framework combining deep learning, machine learning, and time series models for accurate temperature prediction and the detection of severe temperature events. The framework demonstrates superior performance with a Gated Recurrent Unit (GRU) model and incorporates Explainable AI (XAI) to identify key meteorological variables influencing temperature variations.

Objective

• To develop and evaluate a data-driven framework (XAI-TempSpikeDetector) for accurate temperature prediction and the detection of severe temperature spikes (extreme cold events) by integrating deep learning, machine learning, and time series models, while also incorporating explainability strategies.

Study Configuration

• Spatial Scale: Not explicitly defined for the general framework, but the data used is from a specific location (Szeged, Hungary).
• Temporal Scale: Time series forecasting; specific duration and prediction horizon not explicitly defined.

Methodology and Data

• Models used: Linear Regression, Support Vector Machine (SVM), XGBoost, Random Forest, Gated Recurrent Unit (GRU), Long-Short Term Memory (LSTM), Bidirectional LSTM (BiLSTM), Auto Regressive Moving Average (ARIMA), Neural Prophet. SHapley Additive Explanations (SHAP) for explainability.
• Data sources: Weather data from a Kaggle dataset (https://www.kaggle.com/datasets/budincsevity/szeged-weather), which includes variables like temperature, precipitation type, humidity, and wind speed.

Main Results

• The GRU model achieved the highest prediction accuracy among all tested models, with a Mean Squared Error (MSE) of 0.0003, Root Mean Squared Error (RMSE) of 0.0176, Mean Absolute Error (MAE) of 0.0116, and an R-squared (R2) score of 0.9853.
• The framework successfully detected temperature spikes, specifically extreme cold events, which are crucial for early warning systems.
• A SHAP study revealed that the type of precipitation, humidity, and wind speed are the primary variables influencing temperature variations.
• The integration of explainability strategies enhances the detection of severe events while ensuring interpretability and confidence in the predictive models.

Contributions

• Proposal of XAI-TempSpikeDetector, a novel framework that integrates diverse machine learning, deep learning, and time series models for both temperature prediction and severe event detection.
• Identification of the GRU model as the top performer for temperature prediction with high accuracy metrics.
• Successful application of statistical, ML, and DL techniques for detecting extreme cold events, contributing to early warning systems.
• Incorporation of Explainable AI (XAI) using SHAP to provide insights into feature importance, specifically identifying precipitation type, humidity, and wind speed as key drivers of temperature variations.

Funding

• Not explicitly stated in the provided text.

Citation

@article{Krishnan2026XAITempSpikeDetector,
  author = {Krishnan, Archana and Rehiman, K. A. Rafidha and Sabu, M. K.},
  title = {XAITempSpikeDetector: XAI-Powered Temperature Prediction and Spike Detection, A Data-Driven Approach},
  journal = {Lecture notes in networks and systems},
  year = {2026},
  doi = {10.1007/978-3-032-06250-5_28},
  url = {https://doi.org/10.1007/978-3-032-06250-5_28}
}