From Opacity to Clarity: Employing Explainable AI to Interpret CNN Predictions on Winter Road Conditions
Bibliographic record
Abstract
The development of deep learning models for winter road surface conditions (RSC) classification has advanced in recent years. However, most of these models remain nontransparent, limiting confidence in their predictions. This study is a pioneering effort that employs two explainable artificial intelligence (XAI) techniques, gradient-weighted class activation mapping (Grad-CAM) and Shapley additive explanations (SHAP), to clarify the processes through which convolutional neural networks (CNNs) interpret winter RSC imagery. Grad-CAM provides visual explanations by highlighting important regions in the images, while SHAP offers numerical evaluations of feature importance and identifies features with negative contributions. Such scrutiny is vital both practically and methodologically as it offers transparency and facilitates more reliable artificial intelligence (AI) integration into RSC monitoring. Our investigation focuses on three components: CNN’s attention to relevant image features, the influence of training data size, and the impact of varying CNN architectures. The findings demonstrate that CNNs classify RSC imagery by identifying critical features, such as the road surface portion of the image, and that an increased number of training samples enhances learning capabilities. The study further confirms that different architectures might also affect prediction performance. By unveiling the internal decision-making processes of CNNs, this study addresses the transparency gap and contributes to more effective and informed winter road maintenance operations. The integration of XAI techniques ensures that AI models are not only accurate but also interpretable, enhancing their practical applicability. These insights not only reinforce trust in AI’s real-world applications but also deepen understanding of CNNs, thereby encouraging the development of transparent and reliable AI models, particularly for improving winter transportation safety and mobility.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.014 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.002 | 0.003 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.002 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".