× Limited Time Offer ! FLAT 20-40% off - Grab Deal Before It’s Gone. Order Now
Connect With Us
Order Now

ITC573 Data Knowledge and Engineering Assignment Sample

Assignment Brief

Value: 20%
Group Assessment: No
Submission method options: Alternative submission method

TASK

For this task, you are required to download the Airlines dataset which is available in the Massive Online Analysis (MOA) framework. You will need to access the airlines. arff file. Moreover, you are required to download the MOA (https://sourceforge.net/projects/moa- data stream/) software for this task.
Answer the following short answer questions based on the given dataset.

Knowledge obtained through all topics from Topic 1 to Topic 9, particularly Topic 8 and Topic 9 can be useful for assignment help to answering the questions.

Q1. Data Pre-processing [10 marks]

1. Assume that the Airlines dataset has some missing values. Out of the missing value imputation techniques (i.e. Mean Imputation, EMI and DMI) discussed in this subject, which technique do you prefer to use for the Airlines dataset? Why? [5 marks]

2. Again assume that the Airlines dataset has some missing values and some corrupt values. Do you prefer to handle the missing values before handling the corrupt values? Why? [5 marks]

Q2. Incremental Learning [10 marks]

You are required to perform a data mining task to evaluate different incremental learning algorithms. Load the airlines. arff data set into MOA and compare the performance on this data set for the following algorithms:

• HoeffdingTree
• Adaptive Random Forest
• Hoeffding Adaptive Tree
• Leveraging Bagging

Write a response that shows the performance of the different algorithms and comment on their performance using the classification accuracy and other performance metrics used in MOA. In your report consider:

• Is there a difference in performance between the algorithms?
• Which algorithm performs the best?

Your response should include the necessary screenshots, tables, graphs, etc. to make your report understandable to the reader. The recommended word length for the report is 700 to 1000 words.

SUBJECT LEARNING OUTCOMES

This assessment task will assess the following learning outcome/s:

1. be able to compare and critique various data pre-processing techniques.

2. be able to evaluate the usefulness of data cleansing and pre-processing in discovering useful knowledge necessary for critical business decision.

3. be able to evaluate and compare time series data mining approaches for business decision making.

You MUST prepare and present all text answers to the above tasks in a single document file. All images must also be embedded into the document. All files such as your response to the assignment tasks and Turnitin reports will all be in a single directory, identified by your name.

REQUIREMENTS

Your response to each of the questions such as Q1. a and Q1. b should be no longer than 500 words (excluding figures, tables and graphs). Administrative sections of your assessment such as headings, table of contents, reference list and other diagrams and figures are not included in the word count. In text citations are included as part of your word count. Please note that the length of an answer is not very important. The quality and completeness of an answer is important.

For this assessment you’re required to use APA7 referencing to acknowledge the sources that you have used in preparing your assessment. Please visit Referencing at CSU for guidance. In addition, a very useful tool for you to use that demonstrates how to correctly use in-text referencing and the correct way to cite the reference in your reference list is the Academic Referencing Tool (ART).

This assignment must be uploaded to Turnitin. Students will need to download the Turnitin report and submit it via Turnitin along with the assignment. More information on using Turnitin at CSU can be found at https://www.csu.edu.au/current-students/learning-resources/information- planning/assignments/plagiarism-checking

Solution

Introduction

The paper aims to evaluate the data mining models in incremental learning using Massive Online Analysis or MOA tool. In this context, the Airline data has been selected on which the data mining algorithms will be applied after preprocessing it.

Data Pre-processing

Primarily, the discussion about the data pre-processing will be done in this section. One of the most important data preprocessing is the missing data handling without which the data mining algorithm will not respond properly (Tahir & Loo, 2020).

Computation on Missing Values

The data has been collected in ARFF (Attribute Relational File Format) and loaded in MOA. It has not been seen that there are any missing values. However, if the data will contain missing values, there are three ways to handle this and those are as follows:

1. Replace missing values by mean
2. Replace missing values with EMI
3. Replace missing values with DMI

When the EMI (Expected Maximization Imputation) is applied to the data, the missing data will be replaced by the most occurred values in the feature. If this will be done, the central tendency of the data may be changed and for the continuous variable, it will not work properly as the mode is not a valid computation for continuous variables. On the other hand, Decision Tree-Based EMI method is used to replace the missing values by creating the decision tree model and signifying the value at each node (Beaulac & Rosenthal, 2018). So, the DMI method is time-consuming for large datasets. When the missing data is replaced by the mean value of the features, the central tendency is maintained. However, replacing missing values for categorical or nominal features, mean will not work. As airline data contain both the continuous and nominal features, both mean values (for continuous) and the EMI method (for nominal) will have to be used (Islam, 2011).

Computation on Corrupt Values

Corrupt values are not similar to missing values. In most cases, corrupt values come in the form of special characters such as “?” etc. Those characters are not traced when missing values is found as those are not the missing values and rather treated as the valid member of those features with wrong values (Zhu & Wu, 2020). If those values are not replaced, those values will produce the wrong outcome of the data analysis and finally in the application of data mining algorithms. So, those corrupt values in the data need to be removed to make the dataset cleaned and without any kind of the wrong valuation. So, before missing value treatment, the corrupt values need to be converted to missing values or NaN values so that those will be traced at the time of finding the missing values (Ghalib Ahmed Tahir, 2020). So, finally, it will be easier to trace all the missing values and can be replaced by a suitable method.

So, in view of the thought, the corrupt values need to be converted to missing values first and then the imputation of missing values should be done.

Incremental Learning

In this section, the incremental learning methods will be applied to the airline data and the performances will be observed.

Hoeffding Tree

Primarily, the Hoeffding Tree has been applied to the airline data and classified to determine the flight delay (Guo, Wang, Fan, & Li, 2019). The screenshot of the application of this algorithm in MOA is shown below:

Fig-1: Application of Hoeffding Tree

After applying the algorithm, the following outcome has been obtained graphically as follows:

Fig-2: Visualization of Performance for Hoeffding Tree

After applying the Hoeffding Tree, the classification matrics have been stored and enlisted in the following table:

Table-1: Classification Metrics for Hoeffding Tree


Adaptive Random Forest

The Adaptive Random Forest has been applied to the airline data and classified to determine the flight delay. The screenshot of the application of this algorithm in MOA is shown below:


Fig-3: Application of Adaptive Random Forest

After applying the algorithm, the following outcome has been obtained graphically as follows:

Fig-4: Visualization of Performance for Adaptive Random Forest

After applying the Hoeffding Tree, the classification matrics have been stored and enlisted in the following table:

Table-2: Classification Metrics for Adaptive Random Forest

Hoeffding Adaptive Tree

The Hoeffding Adaptive Tree has been applied to the airline data and classified to determine the flight delay (Tahir & Loo, 2020). The screenshot of the application of this algorithm in MOA is shown below:


Fig-5: Application of Hoeffding Adaptive Tree

After applying the algorithm, the following outcome has been obtained graphically as follows:


Fig-6: Visualization of Performance for Hoeffding Tree

After applying the Hoeffding Tree, the classification matrics have been stored and enlisted in the following table:

Table-3: Classification Metrics for Hoeffding Adaptive Tree


Leveraging Bagging

The Leveraging Bagging has been applied to the airline data and classified to determine the flight delay. The screenshot of the application of this algorithm in MOA is shown below:

Fig-7: Application of Leveraging Bagging

After applying the algorithm, the following outcome has been obtained graphically as follows:


Fig-8: Visualization of Performance for Leveraging Bagging

After applying the Hoeffding Tree, the classification matrics have been stored and enlisted in the following table:

Table-4: Classification Metrics for Leveraging Bagging

Finding and Conclusion

The algorithms have been applied to the airline data for the classification of flight delays. While applying the classifiers, the performance matrices have been recorded so that those can be compared to find the best one. From the overall experiment, the following finding has been obtained:

1. The performance of the chosen classifier differs from each other. The accuracy, precision, recall and f1-scores (Overall values and also by class) are different from each of the classifiers that prove the fact that the performances of classification are different from each other.

2. From the comparison of accuracy, it has been seen that the highest accuracy has been obtained from Adaptive Random Forest by 64.3% with the kappa value of 27.33% and precision by 64.04%. The execution time for Adaptive Random Forest (34 seconds) is also comparatively lower with higher accuracy.

References

Beaulac, C., & Rosenthal, J. S. (2018). BEST : A decision tree algorithm that handles missing values. University of Toronto, 1-22.

Ghalib Ahmed Tahir, C. K. (2020). Mitigating Catastrophic Forgetting In Adaptive Class Incremental Extreme Learning Machine Through Neuron Clustering. Systems Man and Cybernetics (SMC) 2020 IEEE International Conference, 3903-3910.

Guo, H., Wang, S., Fan, J., & Li, S. (2019). Learning Automata Based Incremental Learning Method for Deep Neural Networks. IEEE Access, 1-6.

Islam, M. G. (2011). A Decision Tree-based Missing Value Imputation Technique for Data Pre-processing. Proceedings of the 9-th Australasian Data Mining Confere, 41-50.

Tahir, G. A., & Loo, C. K. (2020). An Open-Ended Continual Learning for Food Recognition Using Class Incremental Extreme Learning Machines. IEEE Access, 82328 - 82346.

Zhu, H., & Wu, Y. (2020). Inverse-Free Incremental Learning Algorithms With Reduced Complexity for Regularized Extreme Learning Machine. Access IEEE, 177318-177328.

Fill the form to continue reading

Download Samples PDF

Related Sample

Assignment Services