Date of publication xxxx 00, 0000, date of current version xxxx 00, 0000

[Virtual Presenter] The researchers conducted experiments using a dataset of 1000 samples, which included both positive and negative examples. The data was preprocessed to remove noise and outliers, and then split into training and testing sets. The training set consisted of 800 samples, while the testing set had 200 samples. The researchers used a combination of techniques such as gradient boosting, random forest, and support vector machines to train their models. They employed a technique called cross-validation to evaluate the performance of each model on unseen data. The results showed that the ensemble model outperformed individual models in terms of accuracy and precision. The researchers concluded that the ensemble approach provided better performance than traditional methods, particularly when dealing with complex datasets like this one. The study suggests that ensemble learning can be a valuable tool for improving model performance in various applications..

[Audio] The limitations and challenges of existing approaches to thermal characterization in power electronics are significant. High computational costs associated with machine learning algorithms hinder their implementation in real-time embedded systems. However, recent studies have shown promising results in this area. For example, researchers have successfully deployed SiC MOSFETs in 800-V electric vehicles and emphasized the importance of accurately estimating the peak junction temperature for inverter lifetime prediction. The use of GaN technologies also highlights the need for efficient thermal management. Despite these developments, there are still some limitations in current approaches. Thermal characterization has historically relied on empirical measurements and sensor-based monitoring, which can be time-consuming and costly. Researchers have proposed various solutions to improve accuracy, including multivariate regression models and sensorless estimators. These models enable practical deployment without additional sensors. Virtual sensing techniques have also been introduced, allowing for real-time inference of the peak junction temperature. A compact RC network model has been developed for PV applications, taking into account boundary conditions and thermal coupling effects. Online monitoring methods have been developed for SiC MOSFET modules in machine-drive applications. While these developments have made significant progress in thermal characterization, they are still tailored to specific contexts and primarily focus on regression tasks. A more comprehensive approach is needed that combines thermal risk classification with continuous prediction while maintaining interpretability and computational efficiency. Our hybrid ensemble model addresses these challenges and provides exceptional performance compared to other machine learning algorithms..

[Audio] The hybrid ensemble model combines a voting-based classifier and regressor to predict and classify the peak junction temperature in MOSFETs with high accuracy. The model uses a correlation heatmap to identify the most influential predictors and potential nonlinear dependencies among the input variables. The dataset used for this study includes outputs for the peak Tj in degrees Celsius and the change in junction temperature, referred to as ΔTj. The simulated inputs for the study included two different frequencies, 10 kHz and 15 kHz, and two semiconductor materials, Si and SiC. The dataset provides valuable insights into the aging and reliability of semiconductor devices over time and helps to understand their thermal behavior under variable loads. Maintaining effective management of Tj is critical for enhancing device longevity and performance in power electronics applications such as photovoltaic inverters. Pearson's correlation coefficient was used to quantify the degree of linear association between variables. The hybrid ensemble model will ultimately lead to more accurate predictions and classifications of peak junction temperature in MOSFETs..

[Audio] The predictive architecture developed in this study integrates a voting classifier and a voting regressor into a unified ensemble framework. This framework is designed to simultaneously classify thermal states and provide precise continuous predictions of the junction temperature. The framework leverages the complementary strengths of Random Forest (RF) and Gradient Boosting Machine (XGBoost) to capture both linear and nonlinear feature interactions. The framework also ensures robustness and interpretability..

[Audio] The hybrid ensemble model was designed to predict and classify peak junction temperatures in MOSFETs by combining multiple machine learning models. The model combines two types of classifiers: a voting-based classifier and a voting regressor. The voting-based classifier uses a majority vote approach to make predictions, while the voting regressor uses a weighted average of the predictions made by the base learners. The voting regressor provides continuous estimates of the peak junction temperature, whereas the voting-based classifier provides categorical predictions. Both models were trained on a dataset of MOSFET characteristics and were evaluated using standard classification metrics such as accuracy, precision, recall, and F1-score. The results showed that the hybrid ensemble model performed better than individual models when the data was noisy or had overlapping features. The voting regressor provided more accurate estimates of the peak junction temperature than the voting-based classifier. The model's performance was further evaluated using the Area Under the Receiver Operating Characteristic Curve (ROC-AUC), which measures the model's ability to distinguish between classes. The ROC-AUC values ranged from 0.8 to 1.0, indicating excellent performance. The model's ability to handle mild class imbalance was also evaluated, and the results showed that the hybrid ensemble model performed well even when the classes were imbalanced. The model's robustness was tested by introducing noise into the training data, and the results showed that the model was able to generalize well even in noisy environments. Overall, the hybrid ensemble model demonstrated exceptional performance and reliability in predicting and classifying peak junction temperatures in MOSFETs..

[Audio] The hybrid ensemble model has achieved exceptional performance in predicting and classifying the peak junction temperature in MOSFETs. This is due to its combination of a voting-based classifier and regressor. The model outperforms traditional machine learning algorithms such as support vector machines and k-nearest neighbors. Its precision, recall, and F1-score values are significantly higher than those of these algorithms. The model's ability to handle class imbalance is also noteworthy, as it consistently reaches high precision, recall, and F1-score values for each class. Furthermore, the model demonstrates excellent generalization capabilities across different operating regimes. The macro- and weighted-average scores show close correspondence, indicating the model's reliability and effectiveness. The results suggest that the hybrid ensemble model is an ideal solution for thermal risk prediction in MOSFETs..

[Audio] The proposed voting regressor demonstrates exceptional design of the hybrid ensemble. The soft voting mechanism effectively integrates the complementary strengths of Random Forests (RF), which reduces variance through bootstrap aggregation and XGBoost, which reduces bias via sequential error correction. This synergy enables the model to capture the inherent nonlinearities between Insulation (I), Temperature (Ta) and Thermal Conductivity (Tj). The voting classifier achieves uniformly high F1-scores, minimal confusion-matrix errors, and overall predictive reliability suitable for real-time thermal-risk monitoring of power electronic devices. To further evaluate the generalization behavior of the predictive accuracy, the model achieved a R2 of 0.9951, with a Mean Absolute Error (MAE) of 0.4944 and a Mean Squared Error (MSE) of 1.7554. These results indicate that the model explains nearly all variance in the actual Tj values while maintaining an average absolute deviation of less than half a degree Celsius. Such precision represents an excellent goodness-of-fit for device-level thermal estimation, consistent with the theoretical definition of R2 as the proportion of variance accounted for by the predictive model..

[Audio] The hybrid ensemble model presented here combines a voting-based classifier and regressor to achieve accurate predictions of peak junction temperature in MOSFETs. The model's performance is evaluated based on its mean absolute error (MAE) and residual variance. The results show that the model achieves a sub-degree MAE, indicating excellent agreement between predicted and measured peak junction temperatures. Furthermore, the ensemble's reduction in residual variance relative to individual learners ensures smoother and more consistent temperature estimation across fluctuating irradiance and ambient conditions. The model's convergence behavior and generalization ability are also assessed through residual analysis and learning curve plots. The results demonstrate that the model generalizes effectively and maintains consistent accuracy across the full range of mission profiles. Additionally, the model's explainability is evaluated through SHAP values, which reveal that the dominant predictor of peak junction temperature is the change in junction temperature, followed by current through switching losses and ambient temperature. These findings suggest that the proposed hybrid ensemble model is a valuable tool for thermal reliability management in power-electronic systems..

[Audio] The proposed hybrid ensemble model combines a voting-based classifier and regressor to accurately predict and classify the peak junction temperature in MOSFETs. The model achieves exceptional performance compared to other machine learning algorithms. The ensemble's ability to attribute predictive importance to variables directly tied to heat generation and dissipation mechanisms enhances its interpretability and trustworthiness. The model's learning process is physically grounded, aligning with the physical behavior of semiconductor heating mechanisms. The ensemble's robustness and generalization capability have been demonstrated through its ability to maintain stable learning across multiple cross-validation folds. From an engineering perspective, the proposed hybrid ensemble offers a reliable and interpretable solution for real-time thermal monitoring in power-electronic applications. Its high accuracy and low variance make it particularly suitable for applications where early detection of abnormal temperature rise is essential. The model's performance has been validated through various metrics, including precision, recall, F1-score, and ROC-AUC. The regression component of the model achieved a mean absolute error of 0.4944 and an R-squared of 0.9951, indicating excellent agreement between predicted and measured peak junction temperatures. The most influential predictors of peak junction temperature were found to be ΔTj, Ia, and Ta, while fsb contributed marginally. These results confirm that the model's learning process was physically grounded rather than statistically coincidental. The ensemble's ability to maintain stable learning across multiple cross-validation folds establishes its robustness and generalization capability. The proposed hybrid ensemble model has the potential to bridge the gap between machine learning and thermal physics, paving the way for more intelligent and reliable thermal management systems in next-generation power-electronic applications..

[Audio] The hybrid ensemble model discussed here is a combination of a voting-based classifier and a regressor. This model has been shown to outperform other conventional machine learning algorithms in predicting and classifying peak junction temperature in MOSFETs. The importance of accurately predicting and classifying peak junction temperature in MOSFETs cannot be overstated, particularly in industries such as radiation detection and traction inverters. Several studies have been conducted to develop effective methods for predicting peak junction temperature in MOSFETs. One such study, published in the journal "Nucl. Instrum. Methods Phys. Res." in 2024, used TCAD-augmented machine learning to predict the breakdown performance of SiC PiN diode radiation detectors. The authors of this study, L. Lin, X.-K. Wang, and J. Hu, found success with this approach and published their results under the title "Prediction of JTE breakdown performance in SiC PiN diode radiation detectors using TCAD-augmented machine learning." In 2025, P. Wang and Z. Zhao proposed a gate oxide degradation and junction temperature evaluation method for SiC MOSFETs. This method is based on an on-state resistance model and was published in the journal "Electronics" under the paper number 1. Another study, published in "Sensors" in 2025, developed a junction temperature prediction method that uses multivariate linear regression and the current fall characteristics of SiC MOSFETs. The authors of this study, H. Qin and his colleagues, published their paper under the title "A junction temperature prediction method based on multivariate linear regression using current fall characteristics of SiC MOSFETs." Researchers K. Kong, J. Choi, G. Park, S. Baek, S. Ju, and Y. Han presented a technology for estimating junction temperature in insulated gate bipolar transistors used in traction inverters. This method utilizes a thermal model and was published in "Electronics" in 2025 under paper number 2. In the next study, published in "Sensors" in 2025, Y. Lu and his team proposed a sensorless junction temperature estimation method for onboard SiC MOSFETs using dual-gate-bias-triggered third-quadrant characteristics. The authors of this study, Y. Lu and his team, published their paper under the title "Sensorless junction temperature estimation method for onboard SiC MOSFETs using dual-gate-bias-triggered third-quadrant characteristics." A. Teixeira and his colleagues published a study in the journal "Microelectron. Reliab." in 2023, where they developed a method for estimating the dynamic junction temperature of SiC transistors in order to predict their lifetime in three-phase inverters. This paper, "Precise estimation of dynamic junction temperature of SiC transistors for lifetime prediction of power modules used in three-phase inverters," can be found under paper number 3 in the reference section. Lastly, C. Bianchini, M. Vogni, A. Chini, and G. Franceschini published a paper in 2025 in the journal "Sensors" where they proposed a switching loss model for SiC MOSFETs based on datasheet parameters. This paper can be found under paper number 21 in the reference section. With this, we come to the end of our presentation. We hope you found this information useful and insightful. Thank you for your time and attention..