Neural Classification of Good and Bad Food Using a Feedforward Multi-layer Perceptron with Supervised Learning

Neural networks are an extremely powerful tool for data mining. They are especially useful in cases involving data classification where it is difficult to establish a specific pattern in the search space. In an era when artificial intelligence is increasingly being utilised in industrial and medical applications throughout Africa, it is becoming evident that this is an emerging trend in the continent. Neural networks became popular in the early 20th century and were employed in data classification and pattern recognition. Neural network applications include load forecasting, weather prediction, plant control and time series analysis to mention a few. The concentration of ammonia in cooked food is directly related to its suitability for human consumption. This paper explores the idea of artificial intelligence by employing the use of a feed-forward neural network with two process layers to determine whether a selected cooked food is good or bad. The neural simulation is carried out using Neuro Solutions version 5 software. The food samples tested are yam, rice and beans. The neural network correctly identified the condition of the food samples with an accuracy of 92%. A prototype of the sensor circuit has been constructed for realtime sensing of ammonia concentration in cooked food.