Fruits, including mangoes, produce a wide variety of volatile organic compounds that give them their distinct aroma. Characteristics of fruit aroma is one of the important keys in determining consumer acceptance in the commercial fruit market based on individual preferences. So a possible way to determine the level of ripeness/damage is to feel the distinctive aroma presented by the fruit using a gas sensor. This study aims to build a system that can detect mango damage based on its aroma. The sensors used are TGS 2600, MQ3, MQ4, MQ2, and MQ8 which are connected to the Arduino Mega 2560. The learning model used is an ensemble learning model of Dynamic Classifier Selection (DCS) with Local Class Accuracy (LCA)/DCS-LCA. This algorithm combines Logistic Regression, Selection Tree, Support Vector Machine (SVM), Naïve Bayes, Random Forest, and Neural Networks. The model was then tested with a comparison of the amount of test data and training data of 70%:30%. The test results showed that the overall system Accuracy was 75% and the ability to detect mango fruit damage was 71%. The DCS-LCA classifier model outperforms each of its constituent base classifiers.