Abstract
The performance of artificial neural networks was compared with the performance of discrete choice models in predicting the purchase of products with weak involvement. A comprehensive literature review on the main paradigms of artificial neural networks was carried out, namely variants of the back-propagation algorithm, radial basis function, and genetic computing. Within the class of discrete choice models, the authors restricted the comparison to the multinomial logit model and the mixed logit. The performance of the models was measured in a database of grocery purchases in supermarkets. Artificial neural networks outperformed discrete choice models in predicting sales in supermarkets, and both types of models demonstrated strong predictive power. As a result, both can be reliably used in marketing to estimate individual or collective probabilities of supermarket product purchases.
TopBackground
The literature review is logically divided into two parts and five sections. The first part contains a review of the main paradigms of artificial neural networks. A review of discrete choice models is included in the second part, while the literature review about the comparison of artificial neural networks and discrete choice models is placed in the third part.
Key Terms in this Chapter
Combination Function: Defines the mathematical function that is used to combine the values of the preceding neurons or input variables.
Transfer Function: Consists of two functions, namely a combination function and an activation function.
Error Function: Defines the measure of the difference between the output variables of the ANN, that is, the values produced by the ANN, and the values of the dependent variables in the sample.
Neuron: A neuron performs two operations on input values. The first operation performs computations between the values of the connections and the values received from other neurons. The second operation applies a function to the value resulting from the previous operation and the result of this function is the output value of the neuron.
Learning Algorithm: Defines how the connection weights are changed to learn from the data set.
Architecture of an ANN: It defines how the neurons in a network are organized. The most common architectures are feedforward and recurrent.