\"A credit score is a numerical expression representing the creditworthiness of an individual. A credit score is primarily based on a credit report, information typically sourced from credit bureaus.\"
\n\"Lenders, such as banks and credit card companies, use credit scores to evaluate the potential risk posed by lending money to consumers and to mitigate losses due to bad debt. Lenders use credit scores to determine who qualifies for a loan, at what interest rate, and what credit limits. Lenders also use credit scores to determine which customers are likely to bring in the most revenue.\"
\n\"Credit scoring is not limited to banks. Other organizations, such as mobile phone companies, insurance companies, landlords, and government departments employ the same techniques. Digital finance companies such as online lenders also use alternative data sources to calculate the creditworthiness of borrowers.\"
\n\"Credit scores usually range from 300 to 850 showing the customer's creditworthiness. A customer with a high credit score shows that they are creditworthy and banks will have no problem giving them a loan. If a customer has a low credit score then banks would be hesitant to give out a loan and if they do it might be with a higher interest rate.\"
\n\"There are a number of credit scoring techniques such as hazard rate modeling, reduced form credit models, the weight of evidence models, linear or logistic regression. The primary differences involve the assumptions required about the explanatory variables and the ability to model continuous versus binary outcomes. Some of these techniques are superior to others indirectly estimating the probability of default. Despite much research from academics and industry, no single technique has been proven superior for predicting default in all circumstances.\"
\nOn \"Credit scoring models: Techniques and issues - Author: Nazri Engku
\nJournal of Advanced Research in Business and Management Studies
\nJournal homepage:www.akademiabaru.com/arbms.html
\nISSN: 2462-1935
Artificial intelligence/machine learning based credit scoring model
\n\"Some of the based methods being suggested and explored by researchers are artificial neural networks, genetic algorithms , and artificial immune system.\"
\n\"The techniques used here are broadly called black boxes in the analytics world because interpreting them is difficult. Banks generally use this type of scoring model for upselling or cross-selling different products of a bank to its customers. These techniques usually outperform the statistical-based credit scoring models but fall behind because of their interpretability issues.\"
\n\"Logistic regression is superior to other methods in predicting defaults It is easy to explain, very tractable, convenient, most practical and favorable technique in practice. However, it was suggested that neural network is equally superior as its overall predictive ability is comparativelly high. Nevertheless, logit model produces slightly lower type I error rates i.e. error in classifying bad loan as a good loan with an average of score of 16% compared to 17% for the neural networks.\"
\n\"There are various statistical-based credit scoring model that have been introduced such as linear discriminant analysis , decision trees , Markov chain analysis, probit analysis and logistic regression.\"
\n\"Decision-making trial and evaluation laboratory technique (DEMATEL) can be used to identify the causal-effect relations between the candidate criteria to be included in the credit scoring/risk model. Each decision maker is requested to specify the direct influence between any two criteria based on a scale consisting of 0, 1 ,2, 3, and 4 representing \"no influence\", \"low influence\", \"medium influence\", \"high influence\", and \"very high influence\", respectively. then excute steps that include: calculating the directed influenced matrix normalization and produce the total-relation matrix.\"
\n\"Among the techniques to determine weights are weight-of-evidence and information value technique, ELECTRE and analytic hierarchy process (AHP). Among those techniques, AHP is the most widely used technique. AHP is a technique that simplifies a complex problem by means of hierarchical analysis methodology, which enables subjective judgments among different criteria. One major problem with the AHP process is the consistency of the pairwise comparison matrices. To address this problem a proposed pre-Likert scale-AHP manages to solve the issue.\"
\n\" The statistical-based techniques are still the methods of choice for bankers. Among the techniques, the most popular one is the logistic regression. However, the variables used must be carefully selected and the weights given to the variables must be carefully determined. Unfortunately, the model developed can only be verified with the availability of previous historical data. If the data are not available, then one suitable method of choice to determine the relevant criteria and the suitable weights for the criteria will be through the combination of DEMATEL and pre-Likert scale AHP whereby the judgments will be executed by experts who are directly involved in performing this credit screening task. \"
\nhttps://www.academia.edu/34562267
\nApplication Scorecard ( Used when a customer applies for a new loan.)
Behavioral Scorecard (used in predicting if an existing customer who has a loan is going to default.)
Collection Scorecard (used to predict customers' responses to different strategies for collecting owed money)
\"Credit scorecard models are used to accept or reject the customer's loan application. The customer will be able to see their decision online on the web portal itself. Since most of the decision-making process is now online, many more applications can be processed increasing the turnaround time.\"
\nhttps://en.wikipedia.org/wiki/Credit_scorecards
\nThe FICO score was first introduced in 1989 by FICO, then called Fair, Isaac, and Company. The FICO model is used by the vast majority of banks and credit grantors, and is based on consumer credit files of the three national credit bureaus: Experian, Equifax, and TransUnion. Because a consumer's credit file may contain different information at each of the bureaus, FICO scores can vary depending on which bureau provides the information to FICO to generate the score.\"
\nhttps://en.wikipedia.org/wiki/Credit_score_in_the_United_States#FICO_score
\n\"Credit Karma is an American multinational personal finance company founded in 2007. It has been a brand of Intuit since December 2020. It is best known as a free credit and financial management platform, but its features also include monitoring of unclaimed property databases and a tool to identify and dispute credit report errors. The company operates in the United States, Canada and the United Kingdom.\"
\nhttps://en.wikipedia.org/wiki/Credit_Karma
\nReflections from the biggest fall on Leaderboard (among gold medal contenders)
\nBy Wei Wu - 5 years ago
According to Wei: There is No Shortcut
\nBojan's 1st place with his 5 team-mates (Jahrer, Silogram, RDizzl3, Olivier, Lin)
\nKaggle Competition Teams have even 8 participants! (in the past achieved more than 40 : )
\"A credit score is a numerical expression representing the creditworthiness of an individual. A credit score is primarily based on a credit report, information typically sourced from credit bureaus.\"
\n\"Lenders, such as banks and credit card companies, use credit scores to evaluate the potential risk posed by lending money to consumers and to mitigate losses due to bad debt. Lenders use credit scores to determine who qualifies for a loan, at what interest rate, and what credit limits. Lenders also use credit scores to determine which customers are likely to bring in the most revenue.\"
\n\"Credit scoring is not limited to banks. Other organizations, such as mobile phone companies, insurance companies, landlords, and government departments employ the same techniques. Digital finance companies such as online lenders also use alternative data sources to calculate the creditworthiness of borrowers.\"
\n\"Credit scores usually range from 300 to 850 showing the customer's creditworthiness. A customer with a high credit score shows that they are creditworthy and banks will have no problem giving them a loan. If a customer has a low credit score then banks would be hesitant to give out a loan and if they do it might be with a higher interest rate.\"
\n\"There are a number of credit scoring techniques such as hazard rate modeling, reduced form credit models, the weight of evidence models, linear or logistic regression. The primary differences involve the assumptions required about the explanatory variables and the ability to model continuous versus binary outcomes. Some of these techniques are superior to others indirectly estimating the probability of default. Despite much research from academics and industry, no single technique has been proven superior for predicting default in all circumstances.\"
\nOn \"Credit scoring models: Techniques and issues - Author: Nazri Engku
\nJournal of Advanced Research in Business and Management Studies
\nJournal homepage:www.akademiabaru.com/arbms.html
\nISSN: 2462-1935
Artificial intelligence/machine learning based credit scoring model
\n\"Some of the based methods being suggested and explored by researchers are artificial neural networks, genetic algorithms , and artificial immune system.\"
\n\"The techniques used here are broadly called black boxes in the analytics world because interpreting them is difficult. Banks generally use this type of scoring model for upselling or cross-selling different products of a bank to its customers. These techniques usually outperform the statistical-based credit scoring models but fall behind because of their interpretability issues.\"
\n\"Logistic regression is superior to other methods in predicting defaults It is easy to explain, very tractable, convenient, most practical and favorable technique in practice. However, it was suggested that neural network is equally superior as its overall predictive ability is comparativelly high. Nevertheless, logit model produces slightly lower type I error rates i.e. error in classifying bad loan as a good loan with an average of score of 16% compared to 17% for the neural networks.\"
\n\"There are various statistical-based credit scoring model that have been introduced such as linear discriminant analysis , decision trees , Markov chain analysis, probit analysis and logistic regression.\"
\n\"Decision-making trial and evaluation laboratory technique (DEMATEL) can be used to identify the causal-effect relations between the candidate criteria to be included in the credit scoring/risk model. Each decision maker is requested to specify the direct influence between any two criteria based on a scale consisting of 0, 1 ,2, 3, and 4 representing \"no influence\", \"low influence\", \"medium influence\", \"high influence\", and \"very high influence\", respectively. then excute steps that include: calculating the directed influenced matrix normalization and produce the total-relation matrix.\"
\n\"Among the techniques to determine weights are weight-of-evidence and information value technique, ELECTRE and analytic hierarchy process (AHP). Among those techniques, AHP is the most widely used technique. AHP is a technique that simplifies a complex problem by means of hierarchical analysis methodology, which enables subjective judgments among different criteria. One major problem with the AHP process is the consistency of the pairwise comparison matrices. To address this problem a proposed pre-Likert scale-AHP manages to solve the issue.\"
\n\" The statistical-based techniques are still the methods of choice for bankers. Among the techniques, the most popular one is the logistic regression. However, the variables used must be carefully selected and the weights given to the variables must be carefully determined. Unfortunately, the model developed can only be verified with the availability of previous historical data. If the data are not available, then one suitable method of choice to determine the relevant criteria and the suitable weights for the criteria will be through the combination of DEMATEL and pre-Likert scale AHP whereby the judgments will be executed by experts who are directly involved in performing this credit screening task. \"
\nhttps://www.academia.edu/34562267
\nApplication Scorecard ( Used when a customer applies for a new loan.)
Behavioral Scorecard (used in predicting if an existing customer who has a loan is going to default.)
Collection Scorecard (used to predict customers' responses to different strategies for collecting owed money)
\"Credit scorecard models are used to accept or reject the customer's loan application. The customer will be able to see their decision online on the web portal itself. Since most of the decision-making process is now online, many more applications can be processed increasing the turnaround time.\"
\nhttps://en.wikipedia.org/wiki/Credit_scorecards
\nThe FICO score was first introduced in 1989 by FICO, then called Fair, Isaac, and Company. The FICO model is used by the vast majority of banks and credit grantors, and is based on consumer credit files of the three national credit bureaus: Experian, Equifax, and TransUnion. Because a consumer's credit file may contain different information at each of the bureaus, FICO scores can vary depending on which bureau provides the information to FICO to generate the score.\"
\nhttps://en.wikipedia.org/wiki/Credit_score_in_the_United_States#FICO_score
\n\"Credit Karma is an American multinational personal finance company founded in 2007. It has been a brand of Intuit since December 2020. It is best known as a free credit and financial management platform, but its features also include monitoring of unclaimed property databases and a tool to identify and dispute credit report errors. The company operates in the United States, Canada and the United Kingdom.\"
\nhttps://en.wikipedia.org/wiki/Credit_Karma
\nReflections from the biggest fall on Leaderboard (among gold medal contenders)
\nBy Wei Wu - 5 years ago
According to Wei: There is No Shortcut
\nBojan's 1st place with his 5 team-mates (Jahrer, Silogram, RDizzl3, Olivier, Lin)
\nKaggle Competition Teams have even 8 participants! (in the past achieved more than 40 : )
Hi @mpwolke ! Thank you for the post. I found it interesting to learn about some of the approaches for credit scoring. I agree with you that interpretability of the model is often a requirement in banks, which is why logistic regression models are frequently used. This algorithm appears effective, but improving the F1 metric for a minor class can be challenging when there is a significant class imbalance. Unfortunately, various techniques such as SMOTE, SMOTE-NC, Tomek Links and others cannot be used due to business-customer requirements. Additionally it is crucial to consider how to perform secondary model validation, such as based on PSI metrics and anticipate which features will experience a significant change in PSI during the model creation stage. Through experience, I have concluded that the correct setup of CI/CD/CT/CM processes (MLOps) is more important than the model itself. This is because even the best model will require retraining and it is crucial to detect any drift over time.
", "rawMarkdown": "Hi @mpwolke ! Thank you for the post. I found it interesting to learn about some of the approaches for credit scoring. I agree with you that interpretability of the model is often a requirement in banks, which is why logistic regression models are frequently used. This algorithm appears effective, but improving the F1 metric for a minor class can be challenging when there is a significant class imbalance. Unfortunately, various techniques such as SMOTE, SMOTE-NC, Tomek Links and others cannot be used due to business-customer requirements. Additionally it is crucial to consider how to perform secondary model validation, such as based on PSI metrics and anticipate which features will experience a significant change in PSI during the model creation stage. Through experience, I have concluded that the correct setup of CI/CD/CT/CM processes (MLOps) is more important than the model itself. This is because even the best model will require retraining and it is crucial to detect any drift over time.", "votes": null }, { "id": "2639201", "postDate": "02/06/2024 18:10:18", "content": "Thank you Bratkovsky for explaining so-well those Score approaches. In fact, I 'm a beginner and I haven't any experience on Credit scoring and all the material related to it. Though I've learned a little bit more after reading those articles mentioned on the topic.
", "rawMarkdown": "Thank you Bratkovsky for explaining so-well those Score approaches. In fact, I 'm a beginner and I haven't any experience on Credit scoring and all the material related to it. Though I've learned a little bit more after reading those articles mentioned on the topic.", "votes": null }, { "id": "2641519", "postDate": "02/07/2024 14:34:14", "content": "Thanks @bratkovskyevgeny for such a detailed post. Hopefully would be really beneficial for someone entering this competition for the first time, especially explanations on scorecards, FICO scores and learnings from past competitions.
", "rawMarkdown": "Thanks @bratkovskyevgeny for such a detailed post. Hopefully would be really beneficial for someone entering this competition for the first time, especially explanations on scorecards, FICO scores and learnings from past competitions.", "votes": null }, { "id": "2642806", "postDate": "02/08/2024 12:36:24", "content": "Great job @mpwolke , thank you for sharing. I've been reading Marcos Lopez de Prado's books, and I think you might enjoy them.
", "rawMarkdown": "Great job @mpwolke , thank you for sharing. I've been reading Marcos Lopez de Prado's books, and I think you might enjoy them.", "votes": null }, { "id": "2644374", "postDate": "02/09/2024 12:31:55", "content": "Obrigada pela dica do livro Vitor.
", "rawMarkdown": "Obrigada pela dica do livro Vitor.", "votes": null }, { "id": "3506494", "postDate": "07/31/2026 12:58:33", "content": "Failures of digital transformation rarely, if ever, occur in terms of technology itself. Usually, the system is technically viable but cannot be implemented due to institutional, economic or governance-related constraints. The examples mentioned above serve to illustrate this general tendency.
\nThe software may automate the process, increase transparency and efficiency.
\nIn any case, the ultimate implementation of the project depends on incentives, regulations and willingness of the institution to adapt to changes.
\nA paperless land registry would work, but old-fashioned regulations only facilitate intermediaries by renaming them, without eliminating them. Artificial Intelligence, drones, and digital mapping could identify all the properties that are not being taxed. However, the implementation may be sabotaged due to incentives, contractual obligations or other interests of the stakeholders. In both cases, the technology solves the technological problem. What remains to be addressed is the institutional problem.
\nThis provides valuable lessons for governments and other institutions planning the implementation of digital transformation initiatives. Technology alone does not mean reform; true transformation requires congruence between the technology, legal and institutional context, accountability, incentives and stakeholders. Otherwise, technology simply transforms the previous model into digital one, without any real reform.
", "rawMarkdown": "Failures of digital transformation rarely, if ever, occur in terms of technology itself. Usually, the system is technically viable but cannot be implemented due to institutional, economic or governance-related constraints. The examples mentioned above serve to illustrate this general tendency.\n\nThe software may automate the process, increase transparency and efficiency.\n\nIn any case, the ultimate implementation of the project depends on incentives, regulations and willingness of the institution to adapt to changes.\n\nA paperless land registry would work, but old-fashioned regulations only facilitate intermediaries by renaming them, without eliminating them. Artificial Intelligence, drones, and digital mapping could identify all the properties that are not being taxed. However, the implementation may be sabotaged due to incentives, contractual obligations or other interests of the stakeholders. In both cases, the technology solves the technological problem. What remains to be addressed is the institutional problem.\n\nThis provides valuable lessons for governments and other institutions planning the implementation of digital transformation initiatives. Technology alone does not mean reform; true transformation requires congruence between the technology, legal and institutional context, accountability, incentives and stakeholders. Otherwise, technology simply transforms the previous model into digital one, without any real reform.", "votes": null } ], "comments": [ { "id": 2639184, "author_name": "bratkovskyevgeny", "author_url": "", "post_date": "02/06/2024 17:59:44", "content": "Hi @mpwolke ! Thank you for the post. I found it interesting to learn about some of the approaches for credit scoring. I agree with you that interpretability of the model is often a requirement in banks, which is why logistic regression models are frequently used. This algorithm appears effective, but improving the F1 metric for a minor class can be challenging when there is a significant class imbalance. Unfortunately, various techniques such as SMOTE, SMOTE-NC, Tomek Links and others cannot be used due to business-customer requirements. Additionally it is crucial to consider how to perform secondary model validation, such as based on PSI metrics and anticipate which features will experience a significant change in PSI during the model creation stage. Through experience, I have concluded that the correct setup of CI/CD/CT/CM processes (MLOps) is more important than the model itself. This is because even the best model will require retraining and it is crucial to detect any drift over time.
", "votes": null, "replies": [ { "id": 2639201, "author_name": "mpwolke", "author_url": "", "post_date": "02/06/2024 18:10:18", "content": "Thank you Bratkovsky for explaining so-well those Score approaches. In fact, I 'm a beginner and I haven't any experience on Credit scoring and all the material related to it. Though I've learned a little bit more after reading those articles mentioned on the topic.
", "votes": null, "replies": [] }, { "id": 2641519, "author_name": "ashishsiwach", "author_url": "", "post_date": "02/07/2024 14:34:14", "content": "Thanks @bratkovskyevgeny for such a detailed post. Hopefully would be really beneficial for someone entering this competition for the first time, especially explanations on scorecards, FICO scores and learnings from past competitions.
", "votes": null, "replies": [] } ] }, { "id": 2642806, "author_name": "vitorbl", "author_url": "", "post_date": "02/08/2024 12:36:24", "content": "Great job @mpwolke , thank you for sharing. I've been reading Marcos Lopez de Prado's books, and I think you might enjoy them.
", "votes": null, "replies": [ { "id": 2644374, "author_name": "mpwolke", "author_url": "", "post_date": "02/09/2024 12:31:55", "content": "Obrigada pela dica do livro Vitor.
", "votes": null, "replies": [] } ] }, { "id": 3506494, "author_name": "suman12345677", "author_url": "", "post_date": "07/31/2026 12:58:33", "content": "Failures of digital transformation rarely, if ever, occur in terms of technology itself. Usually, the system is technically viable but cannot be implemented due to institutional, economic or governance-related constraints. The examples mentioned above serve to illustrate this general tendency.
\nThe software may automate the process, increase transparency and efficiency.
\nIn any case, the ultimate implementation of the project depends on incentives, regulations and willingness of the institution to adapt to changes.
\nA paperless land registry would work, but old-fashioned regulations only facilitate intermediaries by renaming them, without eliminating them. Artificial Intelligence, drones, and digital mapping could identify all the properties that are not being taxed. However, the implementation may be sabotaged due to incentives, contractual obligations or other interests of the stakeholders. In both cases, the technology solves the technological problem. What remains to be addressed is the institutional problem.
\nThis provides valuable lessons for governments and other institutions planning the implementation of digital transformation initiatives. Technology alone does not mean reform; true transformation requires congruence between the technology, legal and institutional context, accountability, incentives and stakeholders. Otherwise, technology simply transforms the previous model into digital one, without any real reform.
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "2637663": "#Credit Scorecards and Credit Score\n\n\"A credit score is a numerical expression representing the creditworthiness of an individual. A credit score is primarily based on a credit report, information typically sourced from credit bureaus.\"\n\n\"Lenders, such as banks and credit card companies, use credit scores to evaluate the potential risk posed by lending money to consumers and to mitigate losses due to bad debt. Lenders use credit scores to determine who qualifies for a loan, at what interest rate, and what credit limits. Lenders also use credit scores to determine which customers are likely to bring in the most revenue.\"\n\n\"Credit scoring is not limited to banks. Other organizations, such as mobile phone companies, insurance companies, landlords, and government departments employ the same techniques. Digital finance companies such as online lenders also use alternative data sources to calculate the creditworthiness of borrowers.\"\n\n\"Credit scores usually range from 300 to 850 showing the customer's creditworthiness. A customer with a high credit score shows that they are creditworthy and banks will have no problem giving them a loan. If a customer has a low credit score then banks would be hesitant to give out a loan and if they do it might be with a higher interest rate.\"\n\n#Credit Scoring Techniques\n\n\"There are a number of credit scoring techniques such as hazard rate modeling, reduced form credit models, the weight of evidence models, linear or logistic regression. The primary differences involve the assumptions required about the explanatory variables and the ability to model continuous versus binary outcomes. Some of these techniques are superior to others indirectly estimating the probability of default. Despite much research from academics and industry, no single technique has been proven superior for predicting default in all circumstances.\"\n\n#Artificial intelligence/machine learning based credit scoring model\n\nOn \"Credit scoring models: Techniques and issues - Author: Nazri Engku\n\nJournal of Advanced Research in Business and Management Studies\nJournal homepage:www.akademiabaru.com/arbms.html \nISSN: 2462-1935\n\nArtificial intelligence/machine learning based credit scoring model\n\n\"Some of the based methods being suggested and explored by researchers are artificial neural networks, genetic algorithms , and artificial immune system.\"\n\n\"The techniques used here are broadly called black boxes in the analytics world because interpreting them is difficult. Banks generally use this type of scoring model for upselling or cross-selling different products of a bank to its customers. These techniques usually outperform the statistical-based credit scoring models but fall behind because of their interpretability issues.\"\n\n\"Logistic regression is superior to other methods in predicting defaults It is easy to explain, very tractable, convenient, most practical and favorable technique in practice. However, it was suggested that neural network is equally superior as its overall predictive ability is comparativelly high. Nevertheless, logit model produces slightly lower type I error rates i.e. error in classifying bad loan as a good loan with an average of score of 16% compared to 17% for the neural networks.\"\n\n#Statistical-based Credit Scoring Models\n\n\"There are various statistical-based credit scoring model that have been introduced such as linear discriminant analysis , decision trees , Markov chain analysis, probit analysis and logistic regression.\"\n\n#Decision-Making Trial and Evaluation (DEMATEL)\n\n\"Decision-making trial and evaluation laboratory technique (DEMATEL) can be used to identify the causal-effect relations between the candidate criteria to be included in the credit scoring/risk model. Each decision maker is requested to specify the direct influence between any two criteria based on a scale consisting of 0, 1 ,2, 3, and 4 representing \"no influence\", \"low influence\", \"medium influence\", \"high influence\", and \"very high influence\", respectively. then excute steps that include: calculating the directed influenced matrix normalization and produce the total-relation matrix.\"\n\n#Determining the weight of Criteria using AHP (Analytic Hierarchy Process)\n\n\"Among the techniques to determine weights are weight-of-evidence and information value technique, ELECTRE and analytic hierarchy process (AHP). Among those techniques, AHP is the most widely used technique. AHP is a technique that simplifies a complex problem by means of hierarchical analysis methodology, which enables subjective judgments among different criteria. One major problem with the AHP process is the consistency of the pairwise comparison matrices. To address this problem a proposed pre-Likert scale-AHP manages to solve the issue.\"\n\n\" The statistical-based techniques are still the methods of choice for bankers. Among the techniques, the most popular one is the logistic regression. However, the variables used must be carefully selected and the weights given to the variables must be carefully determined. Unfortunately, the model developed can only be verified with the availability of previous historical data. If the data are not available, then one suitable method of choice to determine the relevant criteria and the suitable weights for the criteria will be through the combination of DEMATEL and pre-Likert scale AHP whereby the judgments will be executed by experts who are directly involved in performing this credit screening task. \"\n\nhttps://www.academia.edu/34562267\n\n#Types of scorecards:\n\n- Application Scorecard ( Used when a customer applies for a new loan.)\n\n- Behavioral Scorecard (used in predicting if an existing customer who has a loan is going to default.)\n\n- Collection Scorecard (used to predict customers' responses to different strategies for collecting owed money)\n\n\"Credit scorecard models are used to accept or reject the customer's loan application. The customer will be able to see their decision online on the web portal itself. Since most of the decision-making process is now online, many more applications can be processed increasing the turnaround time.\"\n\nhttps://en.wikipedia.org/wiki/Credit_scorecards\n\n#FICO score\n\nThe FICO score was first introduced in 1989 by FICO, then called Fair, Isaac, and Company. The FICO model is used by the vast majority of banks and credit grantors, and is based on consumer credit files of the three national credit bureaus: Experian, Equifax, and TransUnion. Because a consumer's credit file may contain different information at each of the bureaus, FICO scores can vary depending on which bureau provides the information to FICO to generate the score.\"\n\nhttps://en.wikipedia.org/wiki/Credit_score_in_the_United_States#FICO_score\n\n#Credit Karma\n\n\"Credit Karma is an American multinational personal finance company founded in 2007. It has been a brand of Intuit since December 2020. It is best known as a free credit and financial management platform, but its features also include monitoring of unclaimed property databases and a tool to identify and dispute credit report errors. The company operates in the United States, Canada and the United Kingdom.\"\n\nhttps://en.wikipedia.org/wiki/Credit_Karma\n\n#To avoid Competition Karma: Most Interesting Topics on 2018 Home Credit default Risk\n\n[Reflections from the biggest fall on Leaderboard (among gold medal contenders)](https://www.kaggle.com/competitions/home-credit-default-risk/discussion/64908)\nBy Wei Wu - 5 years ago\n\nAccording to Wei: There is No Shortcut \n\nBojan's 1st place with his 5 team-mates (Jahrer, Silogram, RDizzl3, Olivier, Lin) \nKaggle Competition Teams have even 8 participants! (in the past achieved more than 40 : )\n\n[Home Aloan](https://www.kaggle.com/competitions/home-credit-default-risk/discussion/64821)", "2639184": "Hi @mpwolke ! Thank you for the post. I found it interesting to learn about some of the approaches for credit scoring. I agree with you that interpretability of the model is often a requirement in banks, which is why logistic regression models are frequently used. This algorithm appears effective, but improving the F1 metric for a minor class can be challenging when there is a significant class imbalance. Unfortunately, various techniques such as SMOTE, SMOTE-NC, Tomek Links and others cannot be used due to business-customer requirements. Additionally it is crucial to consider how to perform secondary model validation, such as based on PSI metrics and anticipate which features will experience a significant change in PSI during the model creation stage. Through experience, I have concluded that the correct setup of CI/CD/CT/CM processes (MLOps) is more important than the model itself. This is because even the best model will require retraining and it is crucial to detect any drift over time.", "2639201": "Thank you Bratkovsky for explaining so-well those Score approaches. In fact, I 'm a beginner and I haven't any experience on Credit scoring and all the material related to it. Though I've learned a little bit more after reading those articles mentioned on the topic.", "2641519": "Thanks @bratkovskyevgeny for such a detailed post. Hopefully would be really beneficial for someone entering this competition for the first time, especially explanations on scorecards, FICO scores and learnings from past competitions.", "2642806": "Great job @mpwolke , thank you for sharing. I've been reading Marcos Lopez de Prado's books, and I think you might enjoy them.", "2644374": "Obrigada pela dica do livro Vitor.", "3506494": "Failures of digital transformation rarely, if ever, occur in terms of technology itself. Usually, the system is technically viable but cannot be implemented due to institutional, economic or governance-related constraints. The examples mentioned above serve to illustrate this general tendency.\n\nThe software may automate the process, increase transparency and efficiency.\n\nIn any case, the ultimate implementation of the project depends on incentives, regulations and willingness of the institution to adapt to changes.\n\nA paperless land registry would work, but old-fashioned regulations only facilitate intermediaries by renaming them, without eliminating them. Artificial Intelligence, drones, and digital mapping could identify all the properties that are not being taxed. However, the implementation may be sabotaged due to incentives, contractual obligations or other interests of the stakeholders. In both cases, the technology solves the technological problem. What remains to be addressed is the institutional problem.\n\nThis provides valuable lessons for governments and other institutions planning the implementation of digital transformation initiatives. Technology alone does not mean reform; true transformation requires congruence between the technology, legal and institutional context, accountability, incentives and stakeholders. Otherwise, technology simply transforms the previous model into digital one, without any real reform." }, "source": "meta" }