German Credit Dataset Research Articles

Research on credit risk assessment optimization based on machine learning

Credit business is a vital part of the bank's core business, which has an extremely important impact on the bank's income and development. In the operation of credit business, credit risk assessment is particularly crucial, and accurate risk assessment can minimize risks while maximizing the bank's returns. We propose a method to optimize credit risk assessment using machine learning techniques. In this work, we employ a random forest machine learning model to process and analyze large amounts of loan application data. By using correlation analysis, information enrichment, etc., the characteristics that have the most impact on credit risk assessment are screened. Subsequently, the model was constructed using a random forest algorithm. Random forests improve the generalization ability and accuracy of the model by building multiple decision trees and introducing randomness between these trees. In the experimental analysis part, we compare the performance of various models on the German credit dataset, and the results show that the deep learning model outperforms the traditional machine learning model in most indicators, verifying the effectiveness of our method.

Optimizing fairness and accuracy: a Pareto optimal approach for decision-making

AbstractIn the era of data-driven decision-making, ensuring fairness and equality in machine learning models has become increasingly crucial. Multiple fairness definitions have been brought forward to evaluate and mitigate unintended fairness-related harms in real-world applications, with little research on addressing their interactions with each other. This paper explores the application of a Minimax Pareto-optimized solution to optimize individual and group fairness at individual and group levels on the Adult Census Income dataset as well as on the German Credit dataset. The objective of training a classification model with a multi-objective loss function is to achieve fair outcomes without compromising utility objectives. We investigate the interplay of different fairness definitions, including definitions of performance consistency and traditional group and individual fairness measures, amongst each other coupled with performance. The results presented in this paper highlight the feasibility of incorporating several fairness considerations into machine learning models, which can be applied to use cases with multiple sensitive features and attributes that characterize real-world applications. This research is a valuable step toward building responsible and transparent machine learning systems that can be incorporated into critical decision-making processes.

Optimizing Ensemble Learning to Reduce Misclassification Costs in Credit Risk Scorecards

Credit risk scorecard models are utilized by lending institutions to optimize decisions on credit approvals. In recent years, ensemble learning has often been deployed to reduce misclassification costs in credit risk scorecards. In this paper, we compared the risk estimation of 26 widely used machine learning algorithms based on commonly used statistical metrics. The best-performing algorithms were then used for model selection in ensemble learning. For the first time, we proposed financial criteria that assess the impact of losses associated with both false positive and false negative predictions to identify optimal ensemble learning. The German Credit Dataset (GCD) is augmented with simulated financial information according to a hypothetical mortgage portfolio observed in UK, European and Australian banks to enable the assessment of losses arising from misclassification costs. The experimental results using the simulated GCD show that the best predictive individual algorithm with the accuracy of 0.87, Gini of 0.88 and Area Under the Receiver Operating Curve of 0.94 was the Generalized Additive Model (GAM). The ensemble learning method with the lowest misclassification cost was the combination of Random Forest (RF) and K-Nearest Neighbors (KNN), totaling USD 417 million in costs (USD 230 for default costs and USD 187 for opportunity costs) compared to the costs of the GAM (USD 487, USD 287 and USD 200). Implementing the proposed financial criteria has led to a significant USD 70 million reduction in misclassification costs derived from a small sample. Thus, the lending institutions’ profit would considerably rise as the number of submitted credit applications for approval increases.

Deep convolutional neural networks with genetic algorithm-based synthetic minority over-sampling technique for improved imbalanced data classification

Imbalanced data classification presents a challenge in machine learning, inducing biased model learning. Moreover, data dimensionality poses another challenge as it highly impacts classifier performance. This paper proposes a new deep-learning method that combines feature selection with oversampling to address these challenges. The proposed approach, GA-SMOTE-DCNN, integrates a genetic algorithm (GA) for feature selection, SMOTE for oversampling, and a deep 1D-convolutional neural network (DCNN) for classification. This study reveals that pre-splitting the data into training and testing sets before applying SMOTE results in higher accuracy, showing an improvement in accuracy ranging between 1.94% and 3.98% compared to post-SMOTE splitting for each dataset. This method achieved accuracy rates of 86.81% for the Balance Scale dataset, 86.15% for the Oil Spill dataset, 89.21% for the Yeast dataset, 91.32% for the Mammography dataset, 88.23% for the Australian credit dataset, and 89.53% for the German Credit dataset when compared with benchmark methods, underscoring its significance in tackling high-dimensional and imbalanced data classification problems. This method demonstrates scalability in effectively addressing challenges associated with high-dimensional and imbalanced data classification across various domains.

On impurity functions in decision trees

Impurity functions are crucial in decision trees. These functions help determine the impurity level of a node in a decision tree, guiding the splitting criteria. However, two primary ambiguities have surrounded impurity functions: (1) the question of their non negativity and (2) the debate over their concavity. In this paper, we address these uncertainties by delving into the characteristics of impurity functions. We establish that the non negativity of an impurity function is inconsequential. Through counter examples, we disprove the equivalence between an impurity function and a concave function. We identify an impurity function that is not concave and a concave function that is not an impurity function. Interestingly, we find an impurity function that results in a negative impurity reduction. Furthermore, we validate several significant properties of impurity functions. For example, we demonstrate that when an impurity function is concave, the impurity reduction remains nonnegative for multiway divisions. We also discuss the sufficient conditions for a concave function to be an impurity function. Our numerical results further indicate that a positive linear combination of the two most popular impurity functions, namely Gini Index and Entropy, may surpass the individual performance of each when applied to the well-known German credit dataset.

Comparative Analysis of Machine Learning Models for Credit Scoring: A Case Study on the South German Credit Dataset

In this research, proposed a novel hybrid machine learning approach that combines the strengths of Random Forest, Multi-Layer Perceptron (MLP), and LightGBM algorithms for classification tasks. This research work focuses on evaluating the performance of this hybrid model using the South German Credit dataset obtained from Kaggle, comprising bank client data, client last contact information, and labels. With 45,211 records and 16 attributes, this dataset provides a suitable environment for assessing the effectiveness of our proposed approach. Employ various evaluation metrics including accuracy, sensitivity and specificity and Receiver Operating Characteristic (ROC) to comprehensively analyze the model's performance. Through experiments, aim to demonstrate the efficacy of the hybrid approach in accurately classifying instances and providing insights into its potential applications in real-world scenarios

Decision Tree C4.5 Performance Improvement using Synthetic Minority Oversampling Technique (SMOTE) and K-Nearest Neighbor for Debtor Eligibility Evaluation

Nowadays, information technology especially machine learning has been used to evaluate the feasibility of debtors. One of the challenges in this classification model is the occurrence of imbalanced datasets, especially in the German Credit Dataset. Another challenge is developing an optimal model for evaluating debtor eligibility. Based on these challenges, this study aims to develop an optimal model for evaluating debtor eligibility on the German Credit Dataset, using the decision trees, k-Nearest Neighbor (k-NN) and Synthetic Minority Oversampling Technique (SMOTE). SMOTE and k-NN is used to overcome challenges regarding imbalanced datasets. While the decision tree are applied to produce a debtor classification model. In general, the steps taken are preparing datasets, pre-processing data, dividing datasets, oversampling with SMOTE, and classification models using decision trees, and testing. Model performance evaluation is represented by accuracy values obtained from the confusion matrix and area under curve (AUC) values generated by the Receiver Operating Characteristic (ROC). Based on the tests that have been carried out, the best accuracy value in the test is obtained at 73.00% and the AUC value is 0.708, in parameters k = 3 and Max-Depth = 25. Based on the analysis produced, the proposed model can improve performance compared to if the dataset is not applied SMOTE.

MINIMIZATION OF RISK AND MAXIMIZATION OF PROFIT ON BEHALF OF BANK BY USING MACHINE LEARNING ALGORITHM

The main aim to develop a loan approval prediction model for a bank by utilizing machine learning algorithms. [Objective] The primary objective is to minimize the bank's loss by developing a decision rule for approving or rejecting a loan application based on an applicant's demographic and socio-economic profile. [Method] Three machine learning algorithms, namely Logistic Regression, Random Forest, and Decision Tree, were applied to a German credit dataset to achieve this objective. The data were preprocessed, including converting the target variable to binary, converting character variables to factors, and selecting only numeric variables. The accuracy, sensitivity, and precision of the models were evaluated using crossvalidation, and the results were compared. [Result] The Random Forest model showed the highest accuracy, with an average accuracy of 76.07%, followed by the Decision Tree model with an average accuracy of 71.99%, and Logistic Regression with an average accuracy of 71.62%. The sensitivity evaluation showed that the Decision Tree model had the highest sensitivity, with an average sensitivity of 70.88%, followed by the Logistic Regression model with an average sensitivity of 68.35%, and the Random Forest model with an average sensitivity of 65.86%.[Conclusion] In conclusion, the Random Forest model showed the highest accuracy, while the Decision Tree model had the highest sensitivity. However, all three models showed similar precision scores. Therefore, based on the objective of minimizing the bank's loss, the Decision Tree model may be more suitable as it has higher sensitivity, which means it has a lower chance of approving a loan for a potentially risky applicant. However, further research and testing are necessary before implementing these models in real-world applications.

Support Vector Machine for Credit System: The Effect of Parameter Optimization and Training Sample

The credit assessment system is an essential part of modern financial institutions, and most of them have adopted different models to perform the task according to their specific needs. Support Vector Machine has been widespread and proved an efficient classifier, especially for relatively small datasets in recent years. When using SVM, data processing, choosing an appropriate kernel function, and tuning parameters can largely affect its performance. The most popular kernel function of SVM is the Radial basis function (RBF), and its main parameters are the regularization parameter, C, and the kernel coefficient, . Our study based on the South German credit dataset demonstrates that parameter optimization and an appropriate ratio of the size of the training dataset to the size of the testing dataset could significantly improve the performance of SVM.

Intelligent credit scoring using deep learning methods

SummaryCredit scoring is one the most important parts of credit risk management in reducing the risk of client defaults and bankruptcies. Deep learning has received much attention in recent years, but it has not been implemented so intensively in credit scoring compared to other financial domains. In this article, stacked unidirectional and bidirectional LSTM (long short‐term memory) networks as a complex area of deep learning are applied in solving credit scoring problems for the first time. The proposed robust model exploits the full potential of the three‐layer stacked LSTM and BDLSTM (bidirectional LSTM) architecture with the treatment and modeling of public datasets in a novel way since credit scoring is not a time sequence problem. Attributes of each loan instance were transformed into a sequence of the matrix with a fixed sliding window approach with a one‐time step. Our proposed models outperform existing and much more complex deep learning solutions thus we succeeded in preserving simplicity. In this article, measures of different types are employed to carry out consistent conclusions. The results by applying three hidden layers on the German Credit dataset showed an accuracy of 87.19%, for Kaggle dataset accuracy reached 93.69%, and for Microcredit dataset accuracy of 97.80%.

Dynamic Effectiveness of Random Forest Algorithm in Financial Credit Risk Management for Improving Output Accuracy and Loan Classification Prediction

With technology impacting several sectors, it can be imagined that the financial sector has a lot to benefit from the increasing level of technological innovations. These institutions take from the surplus of the economy and lend to the deficit sectors of the economy. Individuals and organizations obtain credit facilities from financial institutions to meet basic needs and boost their businesses. However, the stability of the economy is better guaranteed when borrowers pay back the loans availed to them rather than default. This study aims to identify the effectiveness of Random Forest in credit scoring using 32,581 observations. The study proved that Random Forest provides better output accuracy of 91% based on Gini Index for variable selection according to the level of importance when compared to Decision Tree with an output of 83%. It offers better credit scoring accuracy and credit rating as a result of its classification power. The objective of the study is to point out the random forest predictive strength using an unprocessed German credit dataset from Kaggle and to provide an explainable framework sufficient for Financial Institutions and banks to make decisions when granting loans to existing and new applicants.

Seleksi Fitur Menggunakan Eigen Vector Untuk Peningkatan Kinerja K-Means Clustering Dalam Pengelompokan Data

The large number of data set attributes from the data grouping process with K-Means Clustering can affect the number of iterations produced. In this research, Eigen Vector is used to perform feature selection on the data set. The selected data set is then clustered using K-Means Clustering. The data set used in this research is the Wine Quality Dataset obtained from the UCI Machine Learning Repository, with 11 attributes, 4898 data records and 7 attribute classes. Then the South German Credit Dataset was obtained from kaggle.com with 20 attributes, 1000 data records and 2 attribute classes. The results of this research indicate that the number of iterations obtained from the comparison of tests using K-Means without feature selection is that in the Wine Quality Dataset, 11 iterations are obtained, and in the South German Credit Dataset, there are 10 iterations. Meanwhile, K-Means with Eigen Vector feature selection obtained the number of iterations in the Wine Quality Dataset with a total of 5 iterations, and in the South German Credit Dataset with a total of 4 iterations. Clustering evaluation was calculated using Sum of Square Error (SSE). The SSE value in K-Means Clustering without feature selection from the Wine Quality Dataset is 678.5735, while in the South German Credit Dataset it is 1534.3167. While the K-Means Clustering with Eigen Vector from the Wine Quality Dataset is 383.0517, and the South German Credit Dataset is 469.0698. From the results of the proposed method is able to reduce the percentage of errors and minimize the number of iterations on K-Means Clustering with feature selection using Eigen Vector

Development Of Credit Scoring Model For Borrowers Using Machine Learning Techniques

Financial organizations such as banks have experienced an increase in demand for loans from borrowers over the years. These organizations are highly interested in knowing whether a borrower can pay back if granted the loan requested. Granting loans to defaulters can cripple the business, hence, these financial organizations are compelled to evaluate credit worthiness of clients using the vast volume of historical data related to financial position of borrowers. Like other prediction models, credit scoring is a technique used in predicting the probability that a loan applicant, existing borrower, or counterparty will default. Machine learning technique has ability to solve these challenges faced by credit analyst by automating the processing and extraction of knowledge from data. This research focuses on the development of a credit scoring model using Rough Set Theory (RST) and Multi-Layer Perceptron (MLP) Neural Network. RST was used for feature selection while ANN trained with backpropagation was used for classification. This research used two credit scoring datasets; Australian and German credit dataset. Data pre-processing and machine learning were performed using the Anaconda software. This research compares the result obtained from the RST and MLP with Decision Tree, Logistic Regression, Random Forest, Support Vector Machine, Nayes Bayes, K-Nearest Neighbour and ANN using standard evaluation metric to ascertain its performance on the two datasets and conclude the major findings. This research contributes a credit scoring model with improved performance while saving the computational costs.

Bootstrap Bandwidth Selection and Confidence Regions for Double Smoothed Default Probability Estimation

For a fixed time, t, and a horizon time, b, the probability of default (PD) measures the probability that an obligor, that has paid his/her credit until time t, runs into arrears not later that time t+b. This probability is one of the most crucial elements that influences the risk in credits. Previous works have proposed nonparametric estimators for the probability of default derived from Beran’s estimator and a doubly smoothed Beran’s estimator of the conditional survival function for censored data. They have also found asymptotic expressions for the bias and variance of the estimators, but they do not provide any practical way to choose the smoothing parameters involved. In this paper, resampling methods based on bootstrap techniques are proposed to approximate the bandwidths on which Beran and smoothed Beran’s estimators of the PD depend. Bootstrap algorithms for the calculation of confidence regions of the probability of default are also proposed. Extensive simulation studies show the good behavior of the presented algorithms. The bandwidth selector and the confidence region algorithm are applied to a German credit dataset to analyze the probability of default conditional on the credit scoring.

Parameter-Tuned Deep Learning Model for Credit Risk Assessment and Scoring Applications

Background: At present, financial Credit Scoring (CS) is considered as one of the hottest research topics in finance domain, which assists in determining the credit value of individual persons as well as organizations. Data mining approaches are found to be useful in banking sectors, which assist them in designing and developing proper products or services to the customer with minimal risks. Credit risks are linked to loss and loan defaults, which are the main source of risks that exist in the banking sector. Aim: The current research article aims at presenting an effective credit score prediction model for banking sector which can assist them to foresee the credible customers, who have applied for loan. Methods: An optimal Deep Neural Network (DNN)-based framework is employed for credit score data classification using Stacked Autoencoders (SA). Here, SA is applied to extract the features from the dataset. These features are then classified using SoftMax layer. Besides, the network is also tuned Truncated Backpropagation Through Time (TBPTT) model in a supervised way using the training dataset. Results: The proposed model was tested using a benchmark German credit dataset, which includes the necessary variables to determine the credit score of a loan applicant. The presented SADNN model achieved the maximum classification while the model attained high accuracy rate of 96.10%, F-score of 97.25% and kappa value of 90.52%. Conclusion: The experimental results pointed out that a maximum classification performance was attained by the proposed model on all different aspects. The proposed method helped in determining the capability of a borrower in repaying the loan and computing the credit risks properly.

An Automatic Credit Scoring Strategy (ACSS) using memetic evolutionary algorithm and neural architecture search

Credit scoring is playing an increasingly critical role with the rising number of lending operations for micro and small enterprises as well as individuals. A large number of research is primarily based on the combination and construction methods of credit scoring models by the experts. However, different credit data have distinct requirements for the models, and how to automatically search and construct credit scoring models according to credit data has become essential, that is the main concern of this paper. In response to the current challenges for credit scoring research, we proposed an Automatic Credit Scoring Strategy (ACSS), designed a credit assessment platform which includes data import, classification model automatic search, feature selection, hyperparameter optimization, data mining, classification output and other modules. Aiming at the problem of substantial imbalance in credit data, we propose an improved SMOTE algorithm that is capable of generating supplementary data for the lack of minority in credit data, thereby making the credit data distribution well balanced. As for classification model selection, features engineering, parameter optimization and other parts, we further incorporate automatic search ways to reduce manual interaction. We utilize public and self-owned credit data sets to conduct experiments and compare them with the latest credit assessment methods. Extensive experiments have demonstrated that our ACSS method achieves relatively noticeable performance improvements using the German credit dataset, the Taiwan credit dataset and the personal credit dataset for credit scoring. The best results achieved by our proposed method are 0.98, 0.99, 0.895 and 0.901 for MAE, RMSE, accuracy and precision respectively. In addition, the experimental results also show that our proposed improved SMOTE algorithm contributes to the credit scoring performance enhancement. The experimental findings suggest that our proposed ACSS balances automation and accuracy, which can be implemented to the consumption industry to enhance the reliability of credit assessments.

Does Deep Learning Work Well for Categorical Datasets with Mainly Nominal Attributes?

Given the complexity of real-world datasets, it is difficult to present data structures using existing deep learning (DL) models. Most research to date has concentrated on datasets with only one type of attribute: categorical or numerical. Categorical data are common in datasets such as the German (-categorical) credit scoring dataset, which contains numerical, ordinal, and nominal attributes. The heterogeneous structure of this dataset makes very high accuracy difficult to achieve. DL-based methods have achieved high accuracy (99.68%) for the Wisconsin Breast Cancer Dataset, whereas DL-inspired methods have achieved high accuracy (97.39%) for the Australian credit dataset. However, to our knowledge, no such method has been proposed to classify the German credit dataset. This study aimed to provide new insights into the reasons why DL-based and DL-inspired classifiers do not work well for categorical datasets, mainly consisting of nominal attributes. We also discuss the problems associated with using nominal attributes to design high-performance classifiers. Considering the expanded utility of DL, this study's findings should aid in the development of a new type of DL that can handle categorical datasets consisting of mainly nominal attributes, which are commonly used in risk evaluation, finance, banking, and marketing.

Improved Accuracy of Naive Bayes Classifier for Determination of Customer Churn Uses SMOTE and Genetic Algorithms

With increasing competition in the business world, many companies use data mining  techniques to determine the level of customer loyalty. The customer data used in this  study is the german credit dataset obtained from UCI. Such data have an imbalance  problem of class because the amount of data in the loyal class is more than in the  churn class. In addition, there are some irrelevant attributes for customer  classification, so attributes selection is needed to get more accurate classification  results. One classification algorithm is naive bayes. Naive Bayes has been used as an  effective classification for years because it is easy to build and give an independent  attribute into its structure. The purpose of this study is to improve the accuracy of the  Naive Bayes for customer classification. SMOTE and genetic algorithm do for  improving the accuracy. The SMOTE is used to handle class imbalance problems,  while the genetic algorithm is used for attributes selection. Accuracy using the Naive  Bayes is 47.10%, while the mean accuracy results obtained from the Naive Bayes  with the application of the SMOTE is 78.15% and the accuracy obtained from the  Naive Bayes with the application of the SMOTE and genetic algorithm is 78.46%.

KiRTi: A Blockchain-Based Credit Recommender System for Financial Institutions

In this paper, we propose KiRTi , a deep-learning-based credit-recommender scheme for public blockchain to facilitate smart lending operations between prospective borrowers (PB) and prospective lenders (PL) to eliminate the need of third party credit-rating agencies (CRAs) for credit-score (CS) generation. Thus loan grants to PB from PL is secured, authorized, and automated so as to expedite the disbursement process. KiRTi stores PB historical transactions, current assets, and liabilities as time-series sequenced data in a public blockchain. The sequenced data is fetched from blockchain by a long-short term memory (LSTM) model that generates CS for loan recommendations based on proposed lending algorithms for PB and PL. To ensure real-time updation of CS, edge-weights are updated based on boolean indicators from PB and PL, which indicates the successful repayments and loan-defaults. The process is iterated to improve the accuracy of edge-weights and generated CS to ensures the correct credibility of PB for future lending. Smart contracts (SC) are proposed for automatic setup of loan repayments between PB and PL. To model the LSTM recommender scheme, a German credit dataset from UCI repository is considered with 1000 credit-histories of PB, with 700 successful repayments and 300 defaults. KiRTi achieves an accuracy of 97.5% in comparison to conventional approaches with an F-measure of 0.98304. The security evaluation of KiRTi shows that it has computation cost of 20.96 ms and communication cost of 121 bytes compared to other state-of-the-art approaches.

Synthetic Minority Oversampling Technique pada Averaged One Dependence Estimators untuk Klasifikasi Credit Scoring

Bank or financial institution is a business entity whose activities are collecting funds from the public in the form of deposits and channeling them to the public in the form of credit and or other forms. In credit financing problems often occur and one of the problems faced in credit assessment is imbalance class data sets or dataset class imbalances. This problem can be overcome by resampling method, namely by using Oversampling, undersampling and hybrids that combine the two sampling approaches. This research proposes the method of applying SMOTE or Synthetic Minority Oversampling Technique on Averaged One Dependence estimators (AODE) to improve the performance of the accuracy of the credit rating classification on German Credit Creditetsets. The results of this experimental study on the GermanCredit dataset with the classification method without the Resampling process on 13 classifiers produce an average performance value of 70%. The results of the classification with classification techniques that apply the SMOTE method on the AODE algorithm can increase the accuracy performance by 5.5% with an accuracy value of 0.817 or 81.69%. While the classification technique that applies the SpreadSubSample + AODE method decreased by 0.041 or 4.1% but still higher than the accuracy value of other methods with an accuracy value of 0.723 or 72.33%. The researcher concludes that by applying the Resampling technique with the SMOTE method on the AODE algorithm can increase the value of accuracy performance effectively on the imbalance class used for credit scoring or credit rating on GermanCredit datasets.