Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 52 of the 52 references it cites were found in a registry or an index; 2 of the 16 sources its claims cite were checked for support.

Comparative analysis of machine learning approaches for heatwave event prediction in India

Open access
V, Ritesh Choudary, Johnvictor, Anita Christaline, N, Prem Sankar · 2025 · Scientific Reports

Abstract

Heatwaves, are identified as prolonged durations of unusually high temperatures, which pose significant threats to human health, animal health and agriculture. With the increasing frequency and intensity of heatwaves driven by climate change, accurate and early prediction of these extreme weather events is crucial for effective mitigation and adaptation. This research paper conducts a comparative analysis of various machine learning models for heatwave event classification using a time series dataset from a weather station in the equatorial region of India, specifically Chennai, Tamil Nadu. The study evaluates the performance of models including Random Forest, Convolutional Neural Networks, LightGBM, Long Short-Term Memory Networks, Transformer Networks, Support Vector Machines, Graph Neural Networks, Extreme Gradient Boosting and Autoencoders for Anomaly Detection in heatwave. The challenges posed by class imbalance and the limitations of traditional oversampling techniques are discussed, with insights into effective strategies for improving prediction accuracy. Accurate prediction of heatwaves enables mitigation plans to protect humans, animal and plants.

Overview

CoChat Summary

## TL;DR

This paper benchmarks nine machine learning approaches for heatwave event prediction using 2010–2023 weather-station data from Tambaram, Chennai, and finds that CNNs and boosting/ensemble methods perform best. It also argues that SMOTE is unsuitable for this time-series problem because it disrupts temporal structure, and that class weighting is a better imbalance-handling strategy.

## Key findings

  • The dataset covers 14 years of weather observations from Tambaram, Chennai, Tamil Nadu, India and includes 20 meteorological parameters; the study uses a 7-day sliding window and reports an 80/20 train-test split with fivefold cross-validation.
  • CNN achieved the best reported accuracy among the models, with precision 0.73, recall 0.61, F1-score 0.67, ROC AUC 0.9943, and accuracy 0.9892 (98.92%).
  • Random Forest performed strongly with precision 0.80, recall 0.67, F1-score 0.73, ROC AUC 0.9956, and accuracy 0.9912.
  • XGBoost, LightGBM, and Transformer Networks were also among the strongest models, each reported with precision 0.75, recall 0.50, F1-score 0.60, and ROC AUC around 0.993.
  • LSTM performed poorly for heatwave detection, with precision, recall, and F1-score all 0.00 in one reported result set, despite accuracy 0.9824 and ROC AUC 0.9607.
  • SVM and Autoencoders showed high recall but low precision: SVM had precision 0.12, recall 1.00, F1-score 0.22; Autoencoders had precision 0.07, recall 0.94, F1-score 0.13.
  • GNN performed worst, with precision 0.00, recall 0.00, F1-score 0.00 and ROC AUC 0.4714.

## Methods

  • Comparative benchmarking study of nine models: Random Forest, CNN, LightGBM, LSTM, Transformer, SVM, GNN, XGBoost, and Autoencoders for anomaly detection.
  • Data came from a weather station in Tambaram, Chennai; missing values were interpolated, highly correlated features were removed, and feature selection used Pearson correlation plus RandomForestClassifier SelectFromModel.
  • The main analysis used binary classification metrics: precision, recall, accuracy, ROC AUC, and F1-score; deep learning models used binary cross-entropy and Adam, while autoencoders used MSE loss.

## Limitations

  • The paper states that traditional oversampling methods like SMOTE are not suitable for time series because they disrupt temporal dependencies; this is a limitation of the data structure and a motivation for the study’s alternative approach.
  • It also states that CNNs require large training data and are computationally intensive, and that Transformers have high computational cost and need extensive hyperparameter tuning.
  • Evident from the text: the study is based on a single weather-station dataset from one location (Tambaram/Chennai), so the findings may be limited in generalizability beyond this equatorial setting.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s41598-025-04634-9

Cites

52 of 52 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig. 1 Heat prediction workflow implemented in this research.
Fig. 1 Heat prediction workflow implemented in this research.CC BY · PubMed Central
Fig. 2 Temporal and spatial resolution of the dataset with 20 different parameters from Tambaram, Chennai, India.
Fig. 2 Temporal and spatial resolution of the dataset with 20 different parameters from Tambaram, Chennai, India.CC BY · PubMed Central
Fig. 3 Temperature distribution over the 14 years of the dataset.
Fig. 3 Temperature distribution over the 14 years of the dataset.CC BY · PubMed Central
Fig. 4 The number of heat waves in each year.
Fig. 4 The number of heat waves in each year.CC BY · PubMed Central
Fig. 5 Duration of the heatwaves for consecutive days.
Fig. 5 Duration of the heatwaves for consecutive days.CC BY · PubMed Central
Fig. 6 Correlation matrix of features. (a) is Correlation matrix before feature selection. (b) is Correlation matrix after feature selection.
Fig. 6 Correlation matrix of features. (a) is Correlation matrix before feature selection. (b) is Correlation matrix after feature selection.CC BY · PubMed Central
Fig. 7 Radar chart for overall performance.
Fig. 7 Radar chart for overall performance.CC BY · PubMed Central
Fig. 8 Predicted heatwave count for all nine models.
Fig. 8 Predicted heatwave count for all nine models.CC BY · PubMed Central
Fig. 9 Radar chart for overall performance along with persistence prediction model.
Fig. 9 Radar chart for overall performance along with persistence prediction model.CC BY · PubMed Central

Verification

Claims the paper makes

1 Valid · 1 Not valid · 5 Not checked · 53 Own results and interpretations

20 more claims

Create a free account to see all 60 claims and to verify the 3 references still open.

See all claims — free account
  • It may state two separate claims as one. 25
  • Probably not a substantial claim. 12

Create a free account to see these claims.

Citations

Cites

Show all 51 references

Cited by

See all 9

Metadata

Authors
V, R. C., Johnvictor, A. C., & N, P. S.
Affiliations
Vellore Institute of Technology University (IN)
Corresponding author
Johnvictor, Anita Christaline
Venue
Scientific Reports
Published
2025-07-01
Volume
15
Issue
1
Pages
22431-22431
ISSN
2045-2322 (ISSN-L)
OpenAlex
W4411851800
Cited by
9
References
51
Citations by year
2026: 6 · 2025: 3
FWCI
3.58
Type
journal article
Language
en
Open access
Licence
Locations
  • Scientific Reports (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
Vellore Institute of Technology, Chennai
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Anomaly Detection Techniques and Applications
  • Physical Sciences › Environmental Science › Global and Planetary Change › Climate variability and models
  • Physical Sciences › Environmental Science › Health, Toxicology and Mutagenesis › Climate Change and Health Impacts
Keywords
heatwave prediction, extreme weather events, time series forecasting, class imbalance, oversampling, random forest, convolutional neural networks, long short-term memory networks, transformer models, graph neural networks, extreme gradient boosting, autoencoder, anomaly detection, Chennai
MeSH terms
Machine Learning, Humans, India, Neural Networks, Computer, Extreme Heat, Climate Change, Support Vector Machine

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Comparative analysis of machine learning approaches for heatwave event prediction in India. https://app.cochat.ai/papers/10.1038/s41598-025-04634-9
The paper
V, R. C., Johnvictor, A. C., & N, P. S. (2025). Comparative analysis of machine learning approaches for heatwave event prediction in India. Scientific Reports. https://doi.org/10.1038/s41598-025-04634-9

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.