Decision Tree-Based Automatic Curve Model Selection and Comparison with Conventional Curve Fitting Methods
2026 10th International Artificial Intelligence and Data Processing Symposium (IDAP), İstanbul, Türkiye, 5 - 06 Eylül 2026, ss.1-8, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/idap71869.2026.11728061
- Basıldığı Şehir: İstanbul
- Basıldığı Ülke: Türkiye
- Sayfa Sayıları: ss.1-8
- Marmara Üniversitesi Adresli: Evet
Özet
Determining the mathematical model that best describes the input-output relationship of a measurement system is a critical step in sensor characterization and data analysis. With the increasing need for performing data processing directly on measurement devices, automatic model selection has become an important requirement for embedded measurement systems. Although conventional curve fitting tools provide various mathematical models, selecting the most appropriate model generally depends on user experience. Incorrect model selection may reduce characterization accuracy and negatively affect measurement performance. In conventional approaches, linear models are typically estimated using the least squares method, while nonlinear models are optimized using the Gauss–Newton algorithm. The Bayesian Information Criterion (BIC), which considers both regression error and model complexity, is commonly used to select the most suitable model among candidate models. This study proposes a Decision Tree (DT)-based approach for automatically determining the most appropriate mathematical model for sensor measurement data without user intervention. A synthetic dataset containing 10000 measurement curves representing linear, second-order polynomial, third-order polynomial, and exponential characteristics was generated using a calibrated noise model with injected outliers. Fourteen extracted features were used to train the proposed DT model. The proposed method and conventional curve fitting approach were evaluated on 1000 test curves using an STM32F411 microcontroller. The conventional approach achieved 94.8% model selection accuracy with an execution time of 32448.3 µs, whereas the proposed DT-based model achieved 91.5% accuracy with only 406.5 µs execution time. The results demonstrate that the proposed approach provides an effective trade-off between accuracy and computational efficiency, enabling fully automatic curve model selection on Cortex-M4- based embedded measurement devices while reducing processing time by 79.8× compared with conventional methods.