Reference no: EM133698230
Artificial neural network
1. Bankruptcy prediction. Data source: Company Bankruptcy Prediction
1. Sentiment analysis on product/service review. Data source: British Airways Passenger Reviews (2016 - 2023)
2. Once you have decided on a project, download the appropriate dataset from their respective repositories.
3. Go to Science Direct collection at ECU Library. You will need your ECU access credential to log into the database.
4. Search for research articles related to your project. You will use some of them as a foundation for your own analysis. Use the following questions to guide your endeavour. These questions will assist you to make your analysis relevant.
1.
A. What are the findings?
B. Where applicable, what are the variables, including the control variables, they use and why?
C. What are the limitations of these existing studies?
D. How would your analysis extend the existing research?
5. Using appropriate machine learning algorithms and conventional statistical methods, write a report on the followings.
1.
A. The estimators you use in the data analytics. This is the core of your discussion.
I. Which estimators you use for the analysis? E.g., neural network, logistic regression, k-nearest neighbour.
II. Explain why you choose them. What are their strengths and limitations?
III. How well each estimator performs such as their accuracy?
IV. Discuss any limitations in the data, how these affect the estimators' performance, and how you address them.
B. Extract inisghts from the data. Discuss the implications of your findings within a business decision context. Position yourself as an advisor for a group of investors. See the examples below.
I. Bankruptcy prediction: Which metrics (financial ratios) are important and why they are relevant to your client's investment decision.
II. Sentiment analysis: How strong the correlation between the sentiment and the variables of interest? What would be your advice for your clients?
6. Write a 2000-word report on your analysis. The professional report is to be presented to an intelligent, non-specialist audience. You can use these headings to structure your report.
A. Introduction.
B. Methodology.
C. Results, insights, discussion, and recommendation.
D. Limitations and conclusion.
E. References.
Your report is intended for managerial level decision makers. They don't need standardised beta and p-values. They need actionable results. Include persuasive data visualisation where necessary.
Additional information
Introduction
Methodology
Describe the data.
Model specifications such as how many layers used in NN, or how many trees in random forest. This the model architecture. Briefly explain why you choose them. Check prior studies.
What variables you use and why. Check the literature.
How to you assess the model quality? Check the performance metrics below.
Model quality
Metrics: Accuracy, specificity, F1-score, precision, and sensitivity (recall).
Chart: loss function chart (showing training loss and testing loss).
Chart: Area under the curve result.
Results and discussions
Metrics: Accuracy, specificity, F1-score, precision, and sensitivity (recall).
Chart: loss function chart (showing training loss and testing loss).
Chart: Area under the curve result.
Chart: Feature importance chart.
Which algo works best and why?
What variable is the most influential and why?
Limitations
What are the limitations?