The use of coolant in CNC machining for a certain period affects the quality of coolant liquid and requires maintenance and replacement schedules. A coolant that is no longer suitable for use can pollute the surrounding air, water, and soil. The feasibility of using a coolant currently depends on manual observations by the operator. In this research, a sensor system was designed to identify critical change anomalies in coolant in a fast, precise, and nondestructive manner. The system was implemented on a Raspberry Pi equipped with photodiode and gas sensors to identify coolant conditions using IoT and cloud-based machine learning. The use of several gas sensors and one light sensor from this system can obtain different patterns for three coolant quality categories, namely very bad, bad, and good, based on pH testing and microbiological analysis. As new testing data may arrive gradually, this study emphasizes the use of incremental learning algorithms. Our experimental results indicate that the identification of coolant quality has a high success rate, with the highest average accuracy of 96.17% achieved by stochastic gradient descent-logistic regression, followed closely by stochastic gradient descent-SVM at 96.16%, gaussian naive bayes at 94.93%, passive-aggressive at 92.72%, and perceptron at 92.45%. The implementation of this system is expected to replace traditional measurement with the use of human nose and eye senses to obtain the same measurements as different human examiners obtain different results.