Interpretable XGBoost modeling using SHAP and LIME: A real-world electric vehicle study
Öz
Context—As the global automotive market transitions to electric powertrains, the technical determinants of electric vehicle (EV) pricing are still not fully understood, particularly under complex, non-linear interactions among engineering variables. Traditional valuation frameworks tied to mechanical attributes, such as engine displacement and drivetrain layout, are increasingly inadequate for capturing the multidimensional, technology-driven nature of EV market value formation. This limitation underscores the need for data-driven approaches that directly link engineering specifications to economic outcomes.
Objective—This study investigates the key determinants of electric vehicle pricing. By moving beyond vague "black box" predictive models, we established an interpretable machine learning framework. This framework aims to provide stakeholders, from engineers to executives, with clear insight into which specific technical specifications drive market value.
Method—We deployed XGBoost, a high-performance ensemble learning algorithm, on a dataset of 1,137 European EVs. To deconstruct the model’s internal logic, SHAP (Shapley Additive Explanations) and LIME (Local Interpretable Model-agnostic Explanations) were employed. In parallel, Principal Component Analysis (PCA) was applied to consolidate four highly intercorrelated technical variables into a single composite predictor, thereby enabling a system-level representation of vehicle performance rather than an isolated specification-based analysis.
Results—The model explains approximately 96% of the price variance on the held-out test set (RMSE = €6,618; MAE = €3,516). Integrated performance characteristics emerged as the dominant factor, followed by vehicle physical dimensions. Notably, these dimensions exhibit pronounced “threshold effects", where significant price premiums are triggered once specific physical benchmarks are met. In particular, wheelbases above 2,750 mm and electric range plateauing at approximately 450 km. Traditional automotive differentiators, such as drivetrain configuration, had a minimal statistical impact, suggesting that engineering integration and physical specifications are the dominant pricing determinants in the current EV market.
Conclusion—The findings demonstrate that EV market valuation is increasingly shaped by integrated engineering performance and platform-level design parameters rather than legacy mechanical attributes. The study establishes quantitatively grounded design thresholds and shows how interpretable machine learning can translate complex statistical relationships into actionable engineering guidelines. By explicitly connecting model outputs with physical design parameters, this work provides a practical basis for data-driven vehicle development and strategic positioning in the evolving EV landscape.
Anahtar Kelimeler
Destekleyen Kurum
Etik Beyan
Teşekkür
Kaynakça
- A. Recalde, R. Cajo, W. Velasquez, M. S. Alvarez-Alvarado, "Machine learning and optimization in energy management systems for plug-in hybrid electric vehicles: a comprehensive review", Energies, 17(13), 3059, 2024. https://doi.org/10.3390/en17133059.
- I. Demirsoy, "Estimating the Intensity of Point Processes on Linear Networks", Ph.D. dissertation, Florida State University, Florida, USA, 2020.
- A. M. Salih, Z. Raisi-Estabragh, I. B. Galazzo, P. Radeva, S. E. Petersen, K. Lekadir, G. Menegaz, "A Perspective on Explainable Artificial Intelligence Methods: SHAP and LIME", Advanced Intelligent Systems, 7(1), 2400304, 2025. https://doi.org/10.1002/aisy.202400304.
- C. Y. Zhang, S. Cho, M. Vasarhelyi, "Explainable artificial intelligence (XAI) in auditing", International Journal of Accounting Information Systems, 46(SI), 100572, 2022. https://doi.org/10.1016/j.accinf.2022.100572.
- M. Chakraborty, “Explainable Artificial Intelligence (XAI): A Perspective”, Lecture Notes in Networks and Systems, M. Chakraborty, S. P. Chakrabarty, A. Penteado, V. E. Balas, Eds., Singapore. Springer, vol 1148, 2025, 47–63. https://doi.org/10.1007/978-981-97-8457-8_5.
- M. Yildirim, F. Y. Okay, S. Özdemir, "A comparative analysis on the reliability of interpretable machine learning", Pamukkale Üniversitesi Mühendislik Bilimleri Dergisi, 30(4), 494-508, 2023. https://doi.org/10.5505/pajes.2023.49473.
- C. Van Zyl, X. M. Ye, R. Naidoo, "Harnessing eXplainable artificial intelligence for feature selection in time series energy forecasting: A comparative analysis of Grad-CAM and SHAP", Applied Energy, 353, 122079, 2024. https://doi.org/10.1016/j.apenergy.2023.122079.
- A. Gramegna, P. Giudici, "SHAP and LIME: An Evaluation of Discriminative Power in Credit Risk", Frontiers in Artificial Intelligence, 4, 752558, 2021. https://doi.org/10.3389/frai.2021.752558.
Ayrıntılar
Birincil Dil
İngilizce
Konular
Makine Öğrenmesi Algoritmaları, Makine Öğrenme (Diğer), Veri Madenciliği ve Bilgi Keşfi
Bölüm
Araştırma Makalesi
Yazarlar
Idris Demirsoy
*
0000-0002-3321-4748
Türkiye
Erken Görünüm Tarihi
10 Haziran 2026
Yayımlanma Tarihi
-
Gönderilme Tarihi
27 Ocak 2026
Kabul Tarihi
18 Mayıs 2026
Yayımlandığı Sayı
Yıl 2026 Sayı: Advanced Online Publication