Question

Classification and regression are commonly used processes in business analytics. Briefly explain the difference between classification and prediction i. Give examples for classification methods you know. The following diagram shows a neural network with one hidden layer. b1 w1 h1 w5 w2 out w3 i2 w6 h2 W4 b2 Write down the algebraic equation for y, in terms of input values i,i and weights w Briefly explain how neural networks are used for classification iv Give at least three examples how clustering can be used in business analytics. In your answer explain how each business case could be addressed using clustering.

Answer 1

i.

Regression is a modelling technique used when the target variable is continuous; that is, the value we wish to predict/ infer is a continuous value. Examples of this are numbers on a continuous scale (height in inches, rainfall in cm, income, price, etc.)

Classification is a modelling technique used when the target variable is discrete; that is, the value we wish to predict/ infer is a discrete/ class value. For example, predict if an image is a cat or a dog, predict if a transaction is fraud or not fraud.

ii.

There are many classification methods like:

- Logistic regression

- Decision trees

- Random forest

- Naive Bayes classifier

- Neural network

iii.

y1 is a combination of i1 and i2 combined with a combination of weights, and a bias unit added.

Thus, we can say that

h1= b1 + w1*i1 + w3*i2 and

h2= b2 + w2*i1 + w4*i2

and

output,

y1= b3+ w5*h1 + w6*h2

Neural networks are used for classification. The neural networks learn the required weights and biases, which combine the input in different combinations and learn features which can be used to distinguish various classes, thus becoming a powerful mathematical method to perform classification tasks.

iv.

Clustering is a method of unsupervised learning. That is, clustering is used when the target variable is unknown.

Exploratory analysis: Clustering is a powerful method to group similar data points together. This grouping might be difficult to perform manually, but various clustering methods easily calculate distances between data points (both numerical and categorical) and group data together easily, thus helping in exploring and visualizing the data better, and understanding patterns.

Fraud detection: Clustering is also often used to perform fraud detection. Fraud detection is a problem of class imbalance. For every 100 non fraud transactions, there are probably 2 or 3 fraud transactions. Clustering is a powerful tool to group similar non fraud transactions together, and the fraud transactions together, and calculating the distance of a new point from both the classes. This is a very effective method to perform fraud detection.

Customer segmentation: Clustering is also used in customer segmentation, and understanding customer behaviour. Clustering is often used to group similar customers together and performing clear segregations among customer types (high value customers, low value customers, etc).