Uncontrolled release of various harmful gases from automobiles and chemical industries demands accurate methods for gas classification and detection. In this context, this article proposes an effective method to classify and detect four gases - ammonia, formaldehyde, toluene, and acetone using a single field-effect transistor (FET)-based gas sensor. The gate voltage of the FET sensor played a pivotal role in this classification mechanism L-ascorbic acid functionalized graphene oxide (GO) was used as the sensing material of the FET device. Initially, various features of the fabricated FET sensor (i.e., % of response, response time, and recovery time) were captured by varying the applied gate voltage. Furthermore, classification algorithms such as decision tree (DT), support vector machine (SVM), gradient boosting (GB), and random forest (RF) were trained to automatically predict the target gases. An accuracy of 73% was achieved for all three classifiers other than the SVM classifier. The use of machine learning algorithms was fruitful to accurately detect four gases at different gate voltages when any unknown one among the four was exposed to the single gate-tuned sensor. Moreover, it also saved the system's power consumption as a single sensor was behaving like several sensors.

Breath biomarker detection has been a significant non-invasive approach for disease diagnosis. This method has significant potential for early diagnosis and accurate analysis of diseases. Emission from breath contains several volatile organic compounds. Among them, ammonia is a very commonly found VOC and mainly responsible for chronic kidney diseases. There exist several strategies to detect ammonia, however they demonstrate severe limitations such as cross-sensitivity and poor selectivity. This work demonstrates the synergistic effect of sensor functionalization and application of machine learning for selective detection of ammonia in the environment. The sensor exhibits high degree of selectivity towards ammonia owing to enormous hydroxyl groups contributed through curcumin. At 500 ppm ammonia, the sensor demonstrates 274% response and very high selectivity among seven volatile organic compounds. The machine learning models were trained with the help of sensor transients. Random Forest and CNN models were applied to predict the presence of ammonia in a mixture. Random Forest achieved 96.25% accuracy compared to 89% accuracy of CNN. Hence, Random Forest algorithms applied to curcumin functionalized reduced graphene oxide sensors can detect ammonia vapors with very high efficiency among a mixture of gases.