Innovation using hybrid deep neural network detects sensitive ingredients in food products.

Cases of Death of Indonesian Society Due to Poisoning of Food Materials, Especially Packaged Foodstuffs are Assessed Quite High. According to BPOM data, cases of food poisoning are more than 2000 cases per year. This was caused by the lack of literacy received by the community, especially the coastal community about the composition of ingredients in food products. So it takes innovation with in-depth analysis of intelligent system design that can be used by the public to identify certain compositions or ingredients that are visible in the composition table of a product. Barcode With Hybrid Deep Neural Network Method. The aim of this research is to design an intelligent system based on the development of a hybrid deep neural network, convolutional neural network and recurrent neural network, so it is hoped that the system designed from the combination of these methods can provide accurate, effective and efficient detection results. The stages of the research method are [1] Observation and Collection of Barcode Data on Food Composition, Nutritional Value, and Dietary Characteristics, Especially for Coastal Communities. [2] Dataset Preprocessing [3] Deep Neural Network Hybrid Analysis Deepening. [4] Hybrid Deep Neural Network Model Formulation Design Network, [5] Design of Intelligent System for Detection of Sensitive Materials in Food Products [6] Training of Intelligent Hybrid Deep Neural Network System [7] Testing of Intelligent Hybrid Deep Neural Network System in terms of Accuracy, Performance, Efficiency and Effectiveness [8]. Evaluation of Intelligent Systems Using Real Community Data Samples. The results of this study indicate that there are two dominant diseases that are often suffered by people due to food, namely diabetes and hypertension from consuming sugar, salt and flour. From the analysis of the hybrid Deep neural network, which is a combination of CNN and RNN where CNN uses 100 Image data and RNN uses 120 Categorical datasets, it is found from the 40:60 experiment with the accuracy results that 58:42 is 95.23%, 60:40 is 92.50%, 70:30 is 100%, 80:20 is 95% and 90:10 is 100%. Furthermore, for the experiment and RNN, the accuracy of 60:60 is 78.54%. Furthermore, 70:50 is 82.30%, 80:40 is 98.23 %, 50:70 is 98.34%. [ABSTRACT FROM AUTHOR]