Deteksi Cepat Kadar Alkohol Pada Minuman Kopi dengan Metode Dielektrik dan Jaringan Syaraf Tiruan

Penambahan whiskey pada minuman kopi menjadi problem bagi konsumen. Pengukuran perbedaan nilai sifat biolistrik setiap bahan diharapkan dapat memprediksi kadar alkohol dalam minuman kopi. Tujuan penelitian ini adalah untuk memprediksi kadar alkohol dan pH minuman kopi berbasis sifat biolistrik bahan dan Jaringan Syaraf Tiruan (JST). Algoritma backpropagation digunakan menghubungkan input sifat biolistrik dan output prediksi kadar alkohol dan pH minuman kopi ditambah whiskey. Hasil penelitian menunjukkan minuman kopi bersifat kapasitif bahkan resistif. Analisis sensitivitas dengan input sifat biolistrik (induktansi, kapasitansi, resistansi dan impedansi) dan output kadar alkohol dan pH didapat topologi JST terbaik yaitu 4-10-30-2. Pada topologi JST terbaik didapat MSE pelatihan 0,000948 dan MSE validasi 0,0011 serta koefisien korelasi (R) pelatihan sebesar 0,99929 dan R validasi 0,99985. Hasil ini membuka peluang pengembangan teknik deteksi cepat kadar alkohol dalam minuman kopi berbasis sifat biolistrik dengan pemodelan JST. Abstract Whiskey addition in the coffee drinks is a problem for consumers. Measurement of differences in the value of the bioelectric properties of each ingredient is expected to predict alcohol content in coffee drinks. The purpose of this study was to estimate the alcohol content and pH of coffee drinks based on the bioelectric of material and Artificial Neural Networks (ANN). The back propagation algorithm was used to connect the input of bioelectric properties and output of prediction of alcohol content and pH in liqueur coffee. The results showed that liqueur coffee are capacitive and even resistive. Sensitivity analysis with bioelectric properties as input (inductance, capacitance, resistance, and impedance) and alcohol and also pH as output obtained the best ANN topology, 4-10-30-2. The best ANN topology had Mean Standard Error (MSE) of training of 0.000948 and validation MSE of 0.0011 with the correlation coefficient (R) of training and validation of 0.99929 and 0.99985, respectively. These results open up opportunities for the development of rapid alcohol content detection techniques based on bioelectric properties with ANN models for coffee drinks.