etl tools

Kenneth 1

Name: Kenneth Otieno Ouko

Reg no.: COM/1010/17

Course: Neural Networks (COM 424E)

Lecturer: Madam Edna Milgo.

Date: 2/5/2021

CAT 2: 40 marks

1. Compare the similarities of artificial neural network to a biological neural system in

the human brain. (5 marks)

 Both biological neural network and artificial neural network systems process information

in parallel.

 To improve their level of performance, both the networks learn through experience of the

past.

 Learning in both the networks involve adjustment of synaptic connections and weights in

biological neural network and artificial neural network respectively. Weights in ANN is

similar to synaptic connections in the Biological neural system.

 Information transmission in both the networks involve electrical signal.

Kenneth 2

 Storage of information of biological neural network and artificial neural network is at the

synapses and weight matrix respectively.

2. Why is multilayer perception considered suitable to solve non linear problems (the

so called XOR equations)?

When we consider the example of a simple XOR problem. And comparing the abilities of

perceptron and multilayer perceptron in solving the problem,

The diagrams below ascertain the capability of MLP in solving a nonlinear equation and

also give account.

This is an XOR with two inputs x1 and x2.

Pattern space of the XOR

Kenneth 3

Its solution require two lines as shown in the figure below.

Kenneth 4

From this diagram above we can see that the a single line of a perception cannot be used

to solve the problem hence double line is required.

Why is it important to separate training from testing data in supervised models?(5 marks)

This is to avoid overlifting. This is a problem where a model is finely tuned to data it has been

given but fail to give accurate prediction on the data which it has never been trained on.

3. What is a Confusion Matrix.

A confusion matrix is a table that is used to describe the performance of a classification

model (or "classifier") on a set of test data for which the true values are known.

What is Type 1 Error and Type 2 Error in a confusion matrix. Using an example

explain the each of the values in a confusion matrix.(5 marks).

Type I Error (False Positive Error)-is asserting something as true when in reality it is

false. Example, predicting that someone is pregnant she is not. The actual outcome

became false when the prediction was positive.

Type II Error ( False negative): This is when a result of testing shows that a condition

failed while its actual was successfully. This error is committed when we fail to believe a

true condition. Consider the pregnancy example, When we predict that some one isn't

pregnant when in actual sense she is pregnant.

Kenneth 5

4. What is an ROC curve.

Is a graphical representation of the ability of binary classifier, in diagnosis, while its

discrimination threshold varied. Originally, ROC was used in the theory of signal

detection, however its application has extended to medicine, natural hazards, radiology

and machine learning.

Give five advantages of using ROC curve in healthcare data.(5 marks)

 Several diagnostic tasks on a particular subjects could be compared at the same

time in a ROC space.

 By visualizing the curve, sensitivity at a specific FPF can be obtained easily.

 ROC curve analysis can be used to determine optimal cut-off value.

 In contrast to single measures of sensitivity and specificity, the diagnostic

accuracy, such as AUC driven from this analysis is not affected by decision

criterion and it is also independent of prevalence of disease since it is based on

sensitivity and specificity.

5. What is overfitting? What is Underfitting? What happens to an over fitted model

when it is deployed in the real world application.(5 marks)

Overfitting is a situation where a model cannot generalize or fit well on datesets which

are unseen. A sign of overfitting is when the error on the testing dataset is greater than the

error on the training dataset.

Kenneth 6

Underfitting refers to a model that can neither model the training dataset nor generalize

to new dataset.. A Machine learning with under-fitting is not appropriate because it has

poor performance on the training datasets.

6. What is an outliers?

If a data point if considerably far from other data points then its an outlier. Take an

example of a classroom where every student is of an average height except three

extremely tall ones, the three data points, in this case, become outliers.

What do you do when you have skewed data and outliers?

Below are some of the approaches to undertake.

Setup a filter in the testing tool to eliminate skewed and outliers.

Change the values of outliers or skewed data to according to your data.

Remove or change value of outliers during post testing analysis.

Why is it important to normalize data before feeding to a supervised NN.

Normalization of data ensures that, anomalies that may complicate the data analysis are

eliminated. These anomalies may come as a result of deleting data, updating existing

information, and creating more new information.

7. What is deep learning?

Kenneth 7

According to machinelearningmastery.com, deep learning is a subfield of machine

learning concerned with algorithms inspired by the structure and function of the brain

called artificial neural networks.

Why has deep learning gained popularity in the recent years?

Deep learning is superior in giving accurate results on very huge datasets, when trained

Give one example where you would apply deep learning to solve the problem and

why.

Automated Driving: as an automotive researcher,i would use deep learning to detect stop

signs, traffic light and pedestrians on the road.

I would use deep learning because it is superior in predicting outcome of huge

datasets.The system will possibly train itself to give accurate prediction as automation on

road is very crucial.

8. What is convolutional neural network?

Is a kind of algorithm of deep learning that recognizes images by taking different image

inputs, assign weights and bias in various aspects of those images.

How is it different from deep learning and neural network?

How CNN differ from NN- convolutional-neural-network is a subclass of neural-

networks which have at least one convolution layer while CNN, has one or more layers

of convolution units. A convolution unit receives its input from multiple units from the

previous layer which together create a proximity.

Kenneth 8

How CNN differ from deep learning- a convolutional neural network is a class of

deep neural network,commonly applied to analyze visual imagery while deep

learning have several other classes.