Machine learning-guided rational discovery of titanium-doped tungsten oxide gas-sensing materials for enhanced ammonia detection

Tungsten oxide (WO 3 )-based metal oxide semiconductor (MOS) sensors are premier candidates for ammonia (NH 3 ) detection due to their low cost and superior chemical stability. While transition-metal doping effectively modulates sensing performance, identifying the optimal dopant from the vast compositional space remains a trial-and-error bottleneck. Herein, we propose a machine learning (ML)-guided strategy for the rational discovery of high-performance dopants for WO 3 sensors. By leveraging a Random Forest (RF) classifier trained on a rigorously curated dataset, we achieved a high dopant-screening accuracy of 95.8%, identifying titanium (Ti) as the most promising dopant for improving NH 3 response. Guided by the ML predictions, the optimized 3 wt% Ti-doped WO 3 (Ti-WO 3 ) sensor exhibited a response of 16 toward 40 ppm NH 3 at 170 °C, which is approximately 6 times higher than that of pristine WO 3 . Furthermore, the Ti-doped sensor showed markedly accelerated response dynamics, reducing response/recovery time from 170 s / 103 s to 112 s / 45 s, respectively. Moreover, the 3 wt% Ti-WO 3 sensor demonstrated excellent repeatability, high selectivity against interfering gases, and outstanding long-term stability. This work establishes a robust ML-assisted framework for the rational design of advanced gas-sensing materials and offers an efficient route for accelerating the development of high-performance MOS gas sensors.