Week 2: Interdisciplinary Big Data

Huangrui Chu

Economic Meaning

Nowadays, society is developing at a rapid speed, and the total household income of people is increasing, so how can we classify people into different classes like “Rich”, “Medium”, or “Poor”? The first thought that comes into my mind is using Engel’s coefficient(the percentage of income spent on food). However, I need to find my criteria to differentiate people’s Engel’s coefficient. In my decision tree, if a family’s Engel’s coefficient is less than or equal to 0.263, it would say that this family is “Rich”. If a family’s Engel’s coefficient is between 0.263 and 0.535, it would say that this family is “medium”. If a family’s Engel’s coefficient is larger than 0.535, it would say that this family is “Poor”.

In spirits of XAI

Explainable AI is a system that can “explain its capabilities and understandings” (Gunning et al. 2019). My decision tree is not only able to tell me what class a family is in but also tell me how it decides which class a family belongs to by judging the Engel’s coefficient of that family. Thus, my decision tree is an explainable AI approach. In the future, I can feed some other features to train my model and the model can predict the wealth level of a family by considering the job of the family members, the total income, and so on, and it can tell me how it gets this final result.

Ray Zhu

Economic Meaning

My research is aiming at classifying Chinese universities’ budgets into three types, high budget, median budget, and low budget, according to the university type (985, 211, and other universities) and their regions (eastern, middle, and west).
This research can help students and professors predict the amount of a university’s budget and choose institutions that can best support their study and research.
The virtualization tells us that if a Chinese university is a 985 or 211 university, then it has a high budget. But the regional inequality should not be the deciding factor.

In spirits of XAI

Kaiyuan Liu

Economic Meaning

This research intends to analyze how a customer’s annual income and credit score in a bank influence his current loan using a Decision Tree model. For further classification, I uniformly divided the target variable, current loans, into “Low”, “Average” and “High” categories based on the amount, and finally obtained the above visualization. According to the figure, I observe that the current loan amount has a positive correlation with a person’s annual income and credit score. In other words, a customer who has a higher annual income and credit score tends to owe higher loans at present. This observation highly coincides with the real-world economic significance that a customer’s income level and credit rating usually measure his ability to return a loan. From another point, this fact also illustrates that banks are more willing to issue higher loans to those who have better repayment capacity and credit to reduce debiting risks and maximize the profit.

In spirits of XAI

According to the definition, Explainable AI includes a comprehensible system that explains how a conclusion is reached for users (Doran, Schulz, and Besold 2017). This Decision Tree classification exemplifies Explainable AI approaches because it explains how a classification achieves based on what standard and economic factors level by level. Clients, even though they are not data scientists, can comprehend the importance and roles of annual income and credit score in determining a customer’s current loans and understand the reasonability in real-world cases. This application, in my opinion, conceals great potential for banks to make User Profiles. With the help of Explainable AI, banks can precisely recognize whether a customer has default risk and identify an appropriate loan amount for him. In this way, banks not only maximally offer financial aids for their customers but also assume the loss risk to a minimal extent.

Lian Zeng

Economic Meaning

At first, I did a decision tree about the annual transaction amount. The dataset consists of 10,000 credit card clients who mention their age, salary, credit card limit, credit card category, etc. According to the correlation matrix of different effects related to the transaction amount, I selected total transaction counts and credit limits which have a close relationship with the transaction amount. Within the decision tree, the relative institutions can get promoted in those two aspects. For example, adjust the credit limits to fit customers' needs better and promote consumption. Second, I pre-processed the original dataset and chose the average transaction amount and the amount used by customers concerning their credit limit to help predict the "churn" of credit card users. But I found this dataset only has 16.07% of customers who have abandoned credit card services. Therefore, the decision tree I got is relatively unbalanced. However, with a more balanced dataset trained, the business can get insights about carrying out proactive policies capable of providing better services and modifying customer decisions.

In spirits of XAI

XAI aims at providing a more transparent internal system and explanations of their decisions in some level of detail. However, when encountering the huge datasets in real life, in pursuit of applying XAI, the fundamental problem facing explanations of such processing is to find ways to reduce the complexity of all these operations. However, this problem can be solved better by creating a salience map to highlight a small portion of the computation which is most relevant. And the decision trees is such a supervised learning algorithm that can learn the underlying structure of the data and given assumptions and predictions more clearly. (Gilpin et al. 2019 Decision trees can help explainable AI to be more scalable. Instead of using simpler decision algorithms to replace the black-box model which is difficult to be understood, decision trees can help people receive an explanation on automated decisions affecting them and ensure their information privacy for training the black-box model. (Blanco-Justicia et al. 2020

Qunyi Zhao

Economic Meaning

Since the tradeoffs between environmental protection and economic growth has become a huge challenge for China, the research topic I choose is the relationship between AQI (air quality index) and GDP of different cities in China. According to the Decision Tree I made, it predicts that cities with higher AQI (>110.5) are classified into High-Class GDP, whereas cities with low AQI (<=15.5) have lower GDP. This result is understandable: High-GDP cities have more high-tech/energy-powered industries bringing more pollution compared with the labor-intensive industries in low-GDP cities. Based on the output, policymakers can advise to set the warning line of AQI when dealing with rapid industrialization of developed cities. For example, regulators can build plans to reduce coal consumptions, as well as making better filtering system for motor vehicle. This can not only prevent steep economic costs stemming from air pollution, but also raise people’s awareness of environmental degradation.

In spirits of XAI

As Gunning et al. (2019) stated, Explainable AI is a system which gives effective explanations and reasons for each output. The Decision Tree I used just exemplifies XAI approach by presenting the reasons of different classification – using range of AQI for division. The technology enables researchers to better understand XAI and build trust between human and AI at the same time. Future possibility of Explainable AI can bear more social values for generations: By giving explanations for decisions, XAIs can develop more interdisciplinary insights for knowledge application (Gunning et al. 2019). The explanation process can also become credible evidence for policymakers to make regulations for individuals and enterprises. Moreover, it is crucial for XAI to make a qualified balance between accuracy and interpretability in the future.

Bibliography
Gunning, David, Mark Stefik, Jaesik Choi, Timothy Miller, Simone Stumpf, and Guang-Zhong Yang. 2019. “XAI—Explainable Artificial Intelligence.” Science Robotics 4 (37): eaay7120. https://doi.org/10.1126/scirobotics.aay7120.

Ziang Zhou

Economic Meaning

The research topic is to determine the rank of the credit limit for credit card users. The credit limit refers to the maximum balance of your credit card. Our decision tree starts with a utilization ratio of 0.3. By definition, the average utilization ratio measures how heavily you utilize your credit card during a certain period. The users with higher ratios are likely to be ranked as “Average” or “Low” since a higher ratio stands for higher risk of defaulting the payments. However, a lower utilization ratio does not guarantee a higher rank, it also depends on your income category. In an economic website, the author clarifies that the preferred ratio is 30%, which proves our research result convincible (Fiano – n.d.).From the last assignment, we learned that explainable AI is an approach to tackle sophisticated models from mathematical perspectives (Gunning et al. 2019). By managing the amount of each category in the nodes, the decision process is unfolded step by step, where the purity of a node can be measured by entropy or Gini index. Thus, the decision tree classifier is representative of XAI.

In spirits of XAI

From my perspective, it is pivotal to improve the explainability of AI applications for three reasons. First of all, an explainable process offers a shortcut for troubling shooting. Since the process is transparent and trustworthy, we can find the bugs instantly based on the XAI feedback. Also, being able to explain makes AI applications more accessible to the public, and thus attract more people to engage and make contributions to the entire community. Last but not least, by tackling AI in an explainable mathematical approach, the advantage of math in AI could be maximized and thus empower XAI to a great extent.
Bibliography
Fiano –, Liane. n.d. “Credit Score Myths That Might Be Holding You Back from Improving Your Credit.” Consumer Financial Protection Bureau. Accessed January 17, 2021. https://www.consumerfinance.gov/about-us/blog/credit-score-myths-might-be-holding-you-back-improving-your-credit/.
Gunning, David, Mark Stefik, Jaesik Choi, Timothy Miller, Simone Stumpf, and Guang-Zhong Yang. 2019. “XAI—Explainable Artificial Intelligence.” Science Robotics 4 (37): eaay7120. https://doi.org/10.1126/scirobotics.aay7120.

Yifan Song

Economic Meaning

The virtualization is based on the market historical data of real estate valuation collected from New Taipei City, targeting to predict the price of the unit area with features like transaction date and longitude. It allows people to integrate a quantitative model of how the complex input factors affect the price. It is rather economical as AI analysis maximizes the utilization of the data set and delivers an intuitive representation of the price. So people could predict the price using fewer resources and time without losing accuracy. For instance, the diagram shows among other inputs, convenient public transportation (MRT) is the most important feature in real estate valuation. This facilitates price real estates reasonably based on the transportation plus house age or location.

In spirits of XAI

The comprehensive analysis does not need presumptions to identify significant features and it discovers important relations between various inputs. Meanwhile, the decision tree displays which and why features are chosen, giving the system greater “transparency and explainability” (Samek et al. 2019, 6), which makes its behavior more intelligible and convincing. Moreover, displaying the entropy allows users to estimate the system’s reliability and provides information for further improvements. For example, the decision tree is sensitive to noisy data that may cause overfitting, like a node in class low may also contain a number of samples belonging to other classes. XAI not only provides information for the professional but can also help the public understand AI decisions. XAI provides explanations using visualizations, whose output can be interpretable to people without analytical or mathematical backgrounds. For example, the public will be willing and able to utilize the XAI system and learn about real estate valuation.

Xingqi Dai

Economic Meaning

This decision tree predicts the unemployment level of a country according to its business freedom. It shows a negative correlation between business freedom and the unemployment level in a country. Thus, when the government wants to reduce the unemployment rate, they need to improve business freedom and make the business regulatory process more efficient. Then, it becomes easier for citizens to start or operate their own business, and thus more job positions will be created.

In spirits of XAI

Explainable AI provides an explanation to a user who depends on the decisions made by this AI system (Gunning et al. 2019). It can explain what it is doing and disclose the information that it is acting on (Gunning et al. 2019). This decision tree is an explainable system because it explains the numerical criteria for predicting the unemployment level by business freedom, which can be understood by non-technical government officials.
This decision tree shows that explainable AI can be used in policymaking. By collecting data and analyzing them with explainable ai, policymakers can find and understand the relationship between different factors, such as economic freedom or government expenditure, and their political goals, such as reducing the unemployment rate or increasing GDP. Policymakers can also make use of explainable ai to find a way of achieving sustainable development. Explainable ai and big data can tell them if their policy is effective and influential.

Chenjie Yang

Economic Meaning

The virtualization explains the sex feature of the sample determines the decisions made by the tree. We started by three features: sex, alone or not, has cabins or not, but ended up only using sex according to cross-validation on accuracy. By the first split, male samples are in the dead node while female samples in the survived node correspond to Takis's (1999) analysis based on relative frequencies that say 75 percent and 25 percent survival rate for female and male, respectively. Women's survival advantage greatly supports the long believed social norm of "women and children first" (WCF). However, Elinder and Erixson (2012), based on a database of 18 maritime disasters covering over 15,000 individuals' fate, found out that women have a distinct survival disadvantage compared with men. This comparison can evoke research on the ethical dilemma between WCF and "every man for himself" in disaster settings.

In spirits of XAI

In this research, by training the decision tree model, we get an algorithm to predict a passenger's survival on the titanic. The generated virtualization tells how the decision is made with the threshold of each split, the entropy of each node, the number of samples each node has, which perfectly corresponds to Schmelzer's (2019) definition of explainable AI as a tool aiming to explain how machine learning algorithms make black box decisions. The research contains 4 parts, data cleaning, model training, literature review, explaining. The whole process shows explainable AI's potential in hypothesis testing and informing future research. We find our presumption (hypothesis) by data cleaning. Then we train a model to testify it. By literature review and explaining, we can find rationales for either the acceptance or the rejection of our presumption and hence relate our hypothesis to other research to evoke deeper research potentials.

Bibliography
Elinder, M., and O. Erixson. 2012. “Gender, Social Norms, and Survival in Maritime Disasters.” Proceedings of the National Academy of Sciences 109 (33): 13220–24. https://doi.org/10.1073/pnas.1207156109.
Schmelzer, Ron. 2019. “Understanding Explainable AI.” Forbes. July 23, 2019. https://www.forbes.com/sites/cognitiveworld/2019/07/23/understanding-explainable-ai/?sh=6152c5b07c9e.
Takis, Sandra L. 1999. “Titanic: A Statistical Exploration.” The Mathematics Teacher 92 (8): 660–64. https://doi.org/10.5951/mt.92.8.0660.

Haoxin Yu

Economic Meaning

The purpose of my research is to do the online marketing analytics of a store based on the data of 298 customers. The features I choose are Year_birth(birth year), Kidhome(number of children in their household), Teenhome(number of teenagers in their household), Recency(number of days since their last purchase), NumWebVisitsMonth(number of website visit last month).The target is NumWebPurchases(number of online purchases on the website last month) (x), and I set x ≤ 3 as ‘Low’, 3 < x 5 as ‘High’. The results can probably help the store adjust their online marketing strategy and have a general idea of potential user portrait in order to improve the sales amount. For example, they can focus more on the people who have fewer kids, higher recency, younger age, etc.

In spirits of XAI

‘Explainable AI’ refers to humans being able to easily comprehend through dynamically generated graphs or textual descriptions the path artificial intelligence technology took to make a decision (Casey 2019). In this example, the staffs of marketing department or decision maker may not understand even a piece of the algorithms of decision tree. However, they can easily understand the generated graph and adjust the marketing strategies correspondingly, I think this is a process of making AI visualized and comprehensible. From the learning experience, I can get that in many economic activities, to solve a problem is often faced with several different objective conditions or there are several options available. Then, we can use the decision tree and XAI to analyze it, present the results within a graph and find the optimal solution. The process can also be applied in other economic activities like determining a product’s prices, an investment decision-making, etc.

Yihan Zhu

Economic Meaning

The virtualization of the decision tree shows that countries with larger government size have higher economic freedom. Depending on the original complete dataset, countries’ economic freedom is also influenced by property rights, sound money, trade, and regulation. However, we can roughly estimate countries’ economic freedom by their government size according to our decision tree. When people are making policies or economic decisions for countries with a small size of government, they should consider that these countries’ economic freedom may be low, which is related to the system of market and trade there. Therefore, people can’t make similar policies or decisions for them as those countries with large size of government.

In spirits of XAI

Explainable AI (XAI) is where humans can understand its capabilities, causes, process, outcomes, and salient information, and trust its decision by provided explanations (Gunning et al. 2019). Obviously, this hands-on technology (decision tree) does exemplify an XAI approach, since people is easy to understand and trust the relationship between government size and economic freedom of countries from the figure of decision tree based on the clear and accurate dataset.

Ziqiao Ao

Economic Meaning

As is illustrated in the decision tree virtualization, employees enjoy an education wage premium in the U.S. labor market. Specifically, adults who own less than a bachelor's degree are more likely to earn less than 50K, especially those who haven’t completed high school. Contrarily, a master's degree or higher offers employees higher possibilities to earn more than 50K. Based on my prediction of this social issue, implications could be proposed from two perspectives. Firstly, it's worthwhile for jobseekers to improve themselves through higher education while they need to carefully consider the opportunity costs of the tuition and time that could be spent on networking, entrepreneurship, asset investment, etc. (Klazema 2017). Further, under the demand for education, the national government should popularize the education system, especially the nine-year compulsory and university education, and promote the equality of education supply to enhance economic mobility.

In spirits of XAI

According to the definition by Barredo Arrieta et al. (2020), this decision tree model exemplifies an explainable AI approach. On the one hand, it is an intelligent machine learning method that utilizes education to offer a prediction on income automatically with high accuracy. On the other hand, the prediction process is clearly and detailly presented via visualization, including each classification standard, impurity, and sample conditions that could ensure a thorough comprehending without knowing the algorithm behind it. This learning experience consolidates my foregoing opinion that XAI possesses prodigious potential in facilitating interdisciplinary understanding since I could conduct a more effective and targeted literature review on the U.S. income for economic implications based on the classification standards by the decision tree. Besides, analysis of the decision tree model inspires me that explainable AI could also help us assign different levels of importance to elements within factors when making decisions, leading to more efficient time arrangement and unnecessary work deduction.

Xuejun Zhang

Economic Meaning

The dataset is about predicting medical insurance costs based on personal features. Two features with the highest correlation scores are picked - smoker (correlation 0.79) and age (correlation 0.3). The medical insurance cost is divided into 3 levels - low, middle, and high. The result achieves an accuracy score of 0.85 with a decision tree of 2 layers. From the result, we can see that for non-smokers, the younger the person is, the lower the insurance cost. However, for smokers, their insurance cost is high regardless of age. Younger people, especially people under 36, generally spend a lot less money seeing doctors, and smokers cost more medical insurance probably due to deteriorated physical conditions. These results are in line with our common senses, and it provides a more systematical result based on data.

In spirits of XAI

Explainable AI (XAI) is a system that utilizes artificial intelligence technology and can provide reasonable explanations for its behaviors and results (Gunning et al. 2019). This analysis is an example of XAI because it uses AI technology – decision tree, and the result can be explained by analyzing each node of the tree structure. This learning experience provides us with a simple example to build an XAI system. With a simple tree structure, results with accuracy as high as 85% can be achieved. Another AI algorithm that can be used is regression. A regression model can also predict the result with an accuracy of over 95%. It also belongs to the category of explainable AI if we use low-dimensional models. This experiment of the dataset shows us the possibility of using an explainable AI approach to solve economic or social related problems.

Lewis Tian

Economic Meaning

Lucrative industries like finance, consulting, and technology have been famous for their increasingly high recruiting standard especially on the applicants' education, which in part drives the growth of post-graduate education. As NCES (2019) has shown, the number of economic PhDs has increased by over 30% from 2000 to 2016. This visualization harnesses the power of machine learning, specifically a decision tree, to explore the interactions between different criteria and give a sense of what criteria play an important role in the admission process. To maximize gain and minimize opportunity cost, it would be wiser and more effective for students to spend more effort on those aspects that matter more to get a higher chance of admission. As the decision tree shows, GPA and GRE score are picked as the features used to split node. We can interprete that students with CGPA higher 8.845 have a high chance of admission; students who have a GPA <= 8.335 have a low chance of admission, 8.335< GPA <=8.845 have an average chance of admission.

In spirits of XAI

Doran, Schulz, and Besold (2017) describes explainable AI as the AI system that provides not only output but also “human understandable explanation that expresses the rationale of the machine”. And this is a classic explainable AI approach for it utilizes AI technology (decision tree) to provides unique insights in the admission decision, while also produces the explainable visualization so that we can understand how the machine makes such decisions.
Through this hands-on learning experience, XAI is applied to get meaningful insights that might be instrumental for undergraduate students. Also, the idea of learning by practicing is similar to the practicality that XAI embodies. And I firmly believe that the advancement and spread of XAI should be facilitated by actions rather than theories solely. This learning experience also demonstrates the potential of XAI in education. It is more effective to learn AI like the decision tree algorithm, through an XAI application, because students can have a much better grip of how the algorithm works thanks to its explainability.

Sabhyata Jha

Economic Meaning

This decision tree visualization deals with the classification of the purchasing power index of different cities around the world in terms of their property price to income ratio. Purchasing Power Index (PPI) and Property Price to Income Ratio (PPIR) used in the dataset used for this visualization is used as a metric to determine the Quality of Life Index (QLI) of this list of cities. According to Numbeo, the data compiler, PPI and QLI are directly proportional to one another whereas QLI and PPIR are inversely proportional to one another. The property price to income ratio data of all the cites has been broken down into four classes: very high, high, low and very low based on its range. The purchasing power index shows purchasing power in buying goods and services in the given city for the average net salary in that city relative to the purchasing power of New York City.

This exercise is intended to compare the relationship between housing affordability and purchasing power for the possibility of comparing goods and services pricing between countries that have different currencies and economic conditions. There is a correlation between housing prices and a decrease in purchasing power similar to the relationship between inflation and any limited supply of goods (Nguyen 2020). However, the impact of decreasing housing affordability is more severe as it signals a housing crisis and economic inequality as people who are most impacted by it are low-wage and blue-collar workers (Florida 2018). This analysis could also lead to finding a better metric for the market analysis of a geographical location. So, the implications of this analysis could help with research on the financial and economic situation across national economies and, consequently, could be useful for investors, representatives, and professionals of commercial companies, and government bodies for planning and policy purposes.

In spirits of XAI

Decision trees can be comprehended by humans due to their likeness to the human approach by dividing a choice into further sub-choices based on different criteria. This exemplifies an XAI as we can decipher the relationship between the input and output of a supervised learning process (“An Executive’s Guide to AI” 2018). This exercise showed the example of how and what level of human interaction is possible in large-scale integration of analytics on the decision-making level for ethical and optimal use of AI in economics. This experience also showed how following along with the algorithm allows for necessary checks, balances, and modifications that increase the reliability and verifiability of the result, consequently, inspiring the use of XAI as a robust analytics approach.

Xinyu Tian

Economic Meaning

To investigate how the COVID-19 pandemic impacted the life of Chinese people with different types of jobs, researchers collected data from online questionnaires and interviews. Among all the variables of the respondents, gender and the type of job were input as two features to train a decision tree model which analyzed the relationship between personal information and COVID-19 impact degree to people’s life. Separated numbers were used to represent gender and different types of job, while letters A, B, C, and D expressed the impact degree from less to more.

Same as what the raw questionnaire data presented, most people expressed that it affected to a certain extent but within acceptable limits. The tree shows that only the male group with the job in state organs, party-mass organizations, or enterprises and institutions hardly suffered from any impacts. And the males who were students or without employment were heavily affected during the pandemic. Combined with the interview results, it is supposed that male students suffered most from the inconvenience that the quarantine brought and there was a huge decrease in their concentration on study. Hence, this model can be used to predict how much impact the pandemic will make on one person with a certain job, and people can get prepared in advance.

In spirits of XAI

According to the McKinsey & Company, Explainable AI (XAI) is “a set of techniques that help show how a machine-learning algorithm comes up with a set of outputs” (“An executive’s guide to AI,” n.d.). The decision tree exemplified a good XAI approach, for the figure clearly illustrated the algorithm’s pathway in defining the relationship between the input features and outputs. And this indicates that XAI can be developed as a powerful prediction machine that helps in decision making in the future.