A machine learning framework for high-performance capacitive pressure sensor analysis and enhancement

Abstract

Capacitive pressure sensors are widely utilized in industrial, biomedical, and environmental applications due to their high sensitivity, low cost, and compatibility with MEMS technology. However, their accuracy and stability are often compromised by environmental noise, temperature drift, and inherent nonlinearities. This study proposes a machine learning-based compensation method employing Support Vector Machine (SVM) algorithms to improve the performance of high-precision capacitive pressure sensors under realistic conditions. Sensor data were collected under varying temperatures, humidity levels, and electrical noise to simulate practical environments. The raw data was preprocessed and used to train Support Vector Machine (SVM) regression models to correct for nonlinear behavior and drift. The proposed SVM model demonstrated superior performance compared to traditional polynomial and linear calibration techniques. Quantitatively, the root mean square error (RMSE) of the sensor output was reduced by up to 73% after applying the SVM-based compensation. Additionally, the coefficient of determination (R²) increased from 0.84 to 0.97, indicating a significant improvement in prediction accuracy. The proposed model also showed strong generalization when tested on unseen datasets collected under different operating conditions. These findings confirm that machine learning particularly SVM offers a robust and effective solution for real-time error correction in capacitive pressure sensors, paving the way for intelligent, self-calibrating sensing systems with high reliability and precision.