Open Access

ARTICLE

Detecting Vehicle Mechanical Defects Using an Ensemble Deep Learning Model with Mel Frequency Cepstral Coefficients from Acoustic Data

Mudasir Ali¹, Muhammad Faheem Mushtaq², Urooj Akram², Nagwan Abdel Samee^3,*, Mona M. Jamjoom⁴, Imran Ashraf^5,*

1 Department of Computer Science, The Islamia University of Bahawalpur, Bahawalpur, 63100, Pakistan
2 Department of Artificial Intelligence, The Islamia University of Bahawalpur, Bahawalpur, 63100, Pakistan
3 Department of Information Technology, College of Computer and Information Sciences, Princess Nourah bint Abdulrahman University, P.O. Box 84428, Riyadh, 11671, Saudi Arabia
4 Department of Computer Sciences, College of Computer and Information Sciences, Princess Nourah bint Abdulrahman University, P.O. Box 84428, Riyadh, 11671, Saudi Arabia
5 School of Computer Science and Engineering, Yeungnam University, Gyeongsan, 712-749, Republic of Korea

* Corresponding Authors: Nagwan Abdel Samee. Email: ; Imran Ashraf. Email:

Computer Modeling in Engineering & Sciences 2025, 145(2), 1863-1901. https://doi.org/10.32604/cmes.2025.070389

Received 15 July 2025; Accepted 07 October 2025; Issue published 26 November 2025

Abstract

Differentiating between regular and abnormal noises in machine-generated sounds is a crucial but difficult problem. For accurate audio signal classification, suitable and efficient techniques are needed, particularly machine learning approaches for automated classification. Due to the dynamic and diverse representative characteristics of audio data, the probability of achieving high classification accuracy is relatively low and requires further research efforts. This study proposes an ensemble model based on the LeNet and hierarchical attention mechanism (HAM) models with MFCC features to enhance the models’ capacity to handle bias. Additionally, CNNs, bidirectional LSTM (BiLSTM), CRNN, LSTM, capsule network model (CNM), attention mechanism (AM), gated recurrent unit (GRU), ResNet, EfficientNet, and HAM models are implemented for performance comparison. Experiments involving the DCASE2020 dataset reveal that the proposed approach works better than the others, achieving an impressive 99.13% accuracy and 99.56% k-fold cross-validation accuracy. Comparison with state-of-the-art studies further validates this performance. The study’s findings highlight the potential of the proposed approach for accurate fault detection in vehicles, particularly involving the use of acoustic data.

---

Keywords

---