Getting acquainted with neural network analysis

Neural network, popularly known as Artificial Neural Network (ANN) is an information processing system with a large number of nodes and connections as part of a structure which helps in processing complex information. It is influenced by biological human nervous system which consists of a huge number of neurons connected to each other and work together to find solutions for different specified problems. Similarly, ANN sends different responses from different neurons or nodes to the output layer and this outer layer behaves and takes actions accordingly. The collection of neurons or nodes in neural network is structured in three main parts:

  • Input layer,
  • Hidden layer and
  • Output layer.

A short example has been discussed below which will further enlighten ANN process and how it works to achieve a solution.

Working process of neural network

Assuming a company has three departments A, B and C (input layer). The work from all three departments is transferred to managers (hidden layer) and then from managers the work is transferred to director (output layer). Based on the work received the director (output layer) takes a decision. The situation is explained graphically in Figure 1 below.

Figure 1: Working process of neural network

Figure 1: Working process of neural network

Here, the work from A, B and C are transferred to all managers with different productivity. For example, ‘A’ gives 0.2 productivity to ‘X’, 0.5 to ‘Y’ and 0.8 to ‘Z’. Following the same, the total productivity received by X, Y and Z, are 1.4, 0.9 and 1.8 respectively. These productivities are then transferred to directors and total productivity achieved becomes addition of all productivity received from all nodes.

Conducting neural network analysis in SPSS

A bank manager wants to know the factors that may indicate the chances of default on credit card payments. For the same, the manager collects the data of 25 customers about their age, income, credit rating (given by bank) and default history. To perform the neural network analysis, neural network technique in SPSS is selected to know the possible effects or information of ‘neurons’ like age, income, credit rating and default history towards the chances of defaults. To start with neural networks in SPSS:

  1. Select ‘Analyse’
  2. click on ‘Neural Network’
  3. Select ‘Multilayer perception’

Results from neural network in SPSS are shown in various sets like ROC, Variable information, network information, independent variable table and more. For the ease of understanding two main results are explained in this article.

Normalized importance graph

In this graph, all the independent variables are represented with a percentage figure indicating their level of importance in affecting the output. In this case, ‘income’ has been identified as most important variable in suggesting if a credit card customer will default in payment or not. After income, ‘age’ and ‘credit rating’ are important.

Figure 2: Normalized importance of variables

Figure 2: Normalized importance of variables

Neural network diagram

SPSS results also generate neural network diagram of situation. For the present case, SPSS has generated diagram represented in the figure below. Here ‘synaptic weight’ refers to the strength or amplitude of a connection between two nodes, corresponding in biology to the amount of influence the firing of one neuron has on another.

Figure 3: Neural network in SPSS

Figure 3: Neural network in SPSS

As shown in the figure above, income and certain age groups (30-35, 35-40 and 40-45) have more synaptic weight (coefficient estimate). On the other hand, variables like credit rating, history and other age groups have less synaptic weight. As gray connections are indicating ‘positive impact’ or synaptic weight > 0, they all indicate towards ‘YES’ default.

However, blue connections are indicating ‘negative impact’ or synaptic weight < 0 and they all indicate towards ‘NO’ default. Factors like ‘age group’ (25-30, 30-35, 40-45) and ‘income’ have a negative impact on the output layer and that is why it is more inclined towards ‘NO’ default. Other factors are all impacting the output layer in a positive manner and that is why they’re more inclined towards ‘YES’ default.

In conclusion, it can be said that, in this model most of the impact from different factors can be seen on ‘NO’ default. Therefore, this makes the model effective to distinguish and predict the NO default in future customers. These are the factors which will help predicting ‘NO’ defaults most efficiently and the rest will help in explaining ‘YES’ default group of customers.

Application of neural networks

  • Since neural networks are best in identifying patterns or trends in data, they are well suitable for sales forecasting, industrial process control, customer research, data validation, risk management and target marketing.
  • ANN is also useful in recognition of speaker on communications, diagnosis of hepatitis, interpretation of different words, hand writing or facial recognition.
  • Neural network is also important in medical research. It is also used for modeling parts of the human body and recognizing disease from various scans.

Software that support ANN with multiple independent variables are R, SAS, MATLAB, STATA and SPSS.

Priya Chetty

Partner at Project Guru
Priya Chetty writes frequently about advertising, media, marketing and finance. In addition to posting daily to Project Guru Knowledge Tank, she is currently in the editorial board of Research & Analysis wing of Project Guru. She emphasizes more on refined content for Project Guru's various paid services. She has also reviewed about various insights of the social insider by writing articles about what social media means for the media and marketing industries. She has also worked in outdoor media agencies like MPG and hotel marketing companies like CarePlus.

Related articles

  • Understanding random operating curves or ROC analysis Receiver Operating Curve (ROC) is an extension of such classifications. Performance of binary classifier system in the case of ROC analysis can be tested.
  • How to conduct path analysis? Path analysis is a graphical representation of multiple regression models. In this analysis, the graphs represent the relationship between dependent and independent variables with the help of square and arrows.
  • How to apply missing data imputation? Missing data is one of the most common problems in almost all statistical analyses. If the data is not available for all the observations of variables in the model, then it is a case of ‘missing data’. Missing data are part of almost all researches. They are also a common problem in most […]
  • Solving complicated problems with decision tree A decision tree is a graphical representation of possible solutions to a problem based on given conditions. It is called a tree because diagrammatically it starts with a single box (target variable) and ends up in numerous branches and roots (numerous solutions).
  • How to use K-Nearest Neighbor (KNN) algorithm on a dataset? K- Nearest Neighbor, popular as K-Nearest Neighbor (KNN), is an algorithm that helps to assess the properties of a new variable with the help of the properties of existing variables. KNN is applicable in classification as well as regression predictive problems.

Discuss

We are looking for candidates who have completed their master's degree or Ph.D. Click here to know more about our vacancies.