Analisis Respons Sensor Electroni Tongue terhadap Sampel Ganja menggunakan Support Vector Machine

Abstract

Electronic tongue sensors consisting of 16 sensor array made of TOMA and OA lipids that have been used to classify samples of pure cannabis, cannabis mixed with tea and cannabis mixed with tobacco does not involve the feature selection technique so that a lot of duplicated data is generated from data sampling. Feature selection is performed using PCA. Data analysis resulted in loading values shows the contribution of each sensor, and the similarity in sensor performance in characterizing samples, then analyzed using the correlation test so that the sensors that produce redundant information are known. Validation is performed using the SVM method and the classification performance is compared to the original sensor.

The sensor optimization produces a subset of features with 6 sensors (Sensor 7, Sensor 10, Sensor 12, Sensors 13, Sensor 14 and Sensor 15) in the cannabis-tea sample test and a feature subset with 3 sensors (Sensor 3, Sensor 7 and Sensor 14) in the cannabis-tobacco sample test. Sensor optimization that has been done produced classification accuracy by 100% and shorten the running time by a difference of 0.578 microseconds in the test of cannabis-tea samples and a difference of 1.696 microseconds in the test of cannabis-tobacco samples.