Analysis of USDA Food Classifications Using Neural Network Classifier
Neural networks have the potential to analyze many possible situations and learn the correct information through training, learning, and validation. The development of a neural network that is capable of replicating the food groupings in the United States
- PDF / 591,749 Bytes
- 8 Pages / 439.37 x 666.142 pts Page_size
- 14 Downloads / 211 Views
Abstract Neural networks have the potential to analyze many possible situations and learn the correct information through training, learning, and validation. The development of a neural network that is capable of replicating the food groupings in the United States Department of Agriculture’s National Nutrient Database will be analyzed. This database contains 7,906 food items that are grouped into several different categories such as fats and oils, breakfast cereals, and pork products. A single-layer perceptron network is then created to analyze this data and train the network to classify the food items correctly. The neural network data is divided into three categories. Seventy-five percent of the data is used for training the neural network, 15 % is used for validation, and 10 % is used for testing. The results of the network show that certain food groups were harder to classify correctly than others, but overall, the entire testing data set had a misclassification of approximately 10 %. There is still much potential for this network to become better. Expanding the data set in certain categories should help the classification error to decrease even more. Keywords Neural network • Food classification • USDA • Kohonen map • MATLAB • Self-organizing map
1 Introduction The first neural computing model was introduced in the 1940s by McCulloch and Pitts. In the past 70 years, many variations and improvements have been made to this model. This includes adding hidden layers, varying the number of nodes in the hidden layers, changing the activation function of the neurons, implementing momentum to avoid local minima, and developing numerous training methods.
T. Evans • A. Choi (*) Electrical and Computer Engineering, Mercer University, Macon, GA 31207, USA e-mail: [email protected] J. Juang and Y.-C. Huang (eds.), Intelligent Technologies and Engineering Systems, Lecture Notes in Electrical Engineering 234, DOI 10.1007/978-1-4614-6747-2_32, # Springer Science+Business Media New York 2013
259
260
T. Evans and A. Choi
All of these factors influence the performance of the model, but there is no one optimal structure [1–5]. This chapter has concentrated on exploring how these factors influence the performance of a classifier, a neural network used to classify items into predetermined groupings. Artificial neural networks (ANNs) have been used for analyzing many unique cases ranging from analysis to classification. ANNs have been shown to produce better results than most statistical classification techniques. Some previous work completed in the classification area includes analyzing pistachios. Using ANNs, different varieties are capable of being determined by this “computer program” with rather accurate results. By using this program, certain attributes of the pistachio were analyzed and classified based on length and diameter. With this data sample of 5 different varieties, each with 100 samples, the NN was capable of achieving nearly 100 % accuracy. Only 3 pistachios out of 500 were misclassified [6]. Neural networks
Data Loading...
