A fruit dealer has a warehouse to store oranges and bananas. Initially, they are stored together in the same container. To save packaging time and cost, the dealer wants to install an
intelligent machine that can classify the fruits into oranges and bananas automatically. The
fruits are loaded on a conveyor that passes through the machine, which is equipped with
sensors to measure shape, texture, and weight of the fruits. For simplicity, let us assume the
sensors output binary values as follows.
1) Shape: 1 if round; -1 if elliptical
2) Texture: 1 if smooth; -1 if rough
3) Weight: 1 if more than 1 lb; -1 otherwise.
Each fruit is represented by a vector of (Shape Texture Weight]. For training the classifier,
the following instances are provided: [1 -1 -1] for orange (+1) and (-1 1 -1] for banana (-1).
1) Train a perceptron (NOT a MLP) with sign activation function to identify oranges or
bananas using the two instances. Set initial weights W = [0.5 0.5 0.5). Threshold is fixed
at 0.5 (or bias at -0.5) and won't be changed. The learning rate is fixed at 0.5.
2) Test the perceptron with a crushed orange (Shape = -1). De we get the correct target
(orange)? If not, explain briefly why this error occurs and how to improve the classifier.