DETEKSI BERBAGAI JENIS TEH MENGGUNAKAN ELECTRONIC NOSE DENGAN ALGORITMA K-NEAREST NEIGHBORS; DETECTION OF VARIOUS TEA TYPES USING ELECTRONIC NOSE WITH K-NEAREST NEIGHBORS ALGORITHM

Megantoro, Manyang, Danang Lelono

2016 | Skripsi | FMIPA

Electronic nose (eNose) is an instrument that can detect a type of scent using gas sensor array and pattern recognition method. Most gas sensor array used in electronic nose is TGS gas sensor. Actually, it has several drawbacks , one of them is being easily affected by environmental conditions, such as temperature, air pressure and heaping of gas inside the sensor The sensor’s drawbacks can cause the sampled data of scent along some types of tea mixed. As a result, it is difficult to detect the type of tea. Therefore, it is necessary to optimize the data processing, so that the data generated by the eNose still can be used to detect the type of tea. In this research, optimization can be done by widening the distance among the data and choosing a suitable pattern recognition method which can classify new data into the closest group. Widening the range of data is done using noise filter process, baseline manipulation, feature extraction and dimension reduction (PCA). Meanwhile, for grouping the new data (classification ), it uses k-NN pattern recognition method. The results show that the average accuracy achieved by k-nn is 52.40% using Canberra distance search method with the value of k = 5. Whereas, the highest accuracy is obtained on testing a green tea using Canberra distance search method with the value of k = 5 that the accuracy is equal to 85.31% . For type of tea which most widely selected at the time of testing is green tea, with a maximum total of 323 elected. Regarding the results, it can be concluded that the best data processing is by using several steps, namely noise filter, feature extraction, dimension reduction and k-nn without baseline manipulation.

Kata Kunci : noise filter, baseline manipulation, feature extraction, PCA.