Introduction to Neural Networks and its application in Predictive Analytics
What is a neural network?
A neural network is a computational model that has been designed using the concept of a human brain. A human brain is actually a huge network of neurons that store information absorbed by a human. Similarly, a neural network is an interconnected network of nodes that store information and patterns.
The aim of the researchers who came up with the idea of neural networks was to create a system that works like a human brain. The ways in which a neural network resembles the human brain are:
1. It learns with new data inputs and new patterns, hence it is an intelligent system.
2. Similar to the human brain where information is stored in nodes, neural network stores information in the inter-neuron space known as the synaptic weights.
A Neural networks can work on both linear and non-linear data and hence are more useful than the linear predictors used before. They can be used to recognise complex patterns that are often non-linear in order to predict an outcome.
The learning capability of a neural network can very effectively be used for predictions. Neural Networks consists of 3 layers – input, hidden layer (nodes) and output. The hidden layer consists of nodes actually responsible for the output. Output is a function of the weight associated with a node and the input. Considering a simple neural network consisting of two inputs and one output, the output initially is user decided. The system is then given multiple inputs and the output given by the neural network is compared with the one given by the user. This is repeated until output given by the system is exactly equal to the one given by the user. This process is called training a neural network. The system can now be used to predict outputs based on the training that it has received
Application in Predictive Analytics
1. Function approximation – Considering the above example with two inputs and an output decided by the user, it can be made like function approximation. Through repetitive correction, the nodes adjusted their weight in order to get closer to the value given by the user. Because of their ability to constantly learn, the neural networks can do more complex and accurate approximations in comparison to the human brain
2. Forecasting – Neural Networks can be trained with specific data to predict an outcome. For example, a network can be fed with the weather forecast of the last 20 weeks of New York City and then be asked to predict the forecast of the next week. This forecast technique is not just limited to the weather but finds a wide application in the market. Different kinds of data can be fed to the neural network in order to train it to give an accurate forecast
3. Classification – Data classification is one of the most tedious and tricky tasks for humans. A neural network can be trained to classify data on the basis of its training. For example, a network can be fed with multiple loan applications and based on their credentials be asked to categorize them into loan-worthy and not loan-worthy. The classification credentials can also be decided by the user and this quality makes neural networks a very flexible commodity
4. Clustering – Clustering is a more complex form of classification where the number of categories are not pre known. A neural network can be trained by keeping the number of outputs in a particular category as a parameter to form a cluster. Clustering is a very tedious task and hence the neural networks can be used to make it a fast process