This dashboard communicates the ABZ business Loan Default Analysis for loans issued from 2007 to 2013:

• The Metadata Info pane shows the data structure, before and after cleaning the data set, it shows visualization of missing values, data distribution and major statistics to look at before deciding on how to further manipulate/ analyse the data set. A clean data function was written to handle the data cleaning and other feature engineering.

• The Loan Overview pane shows various information into current loan status and general loan status which is formulated to contain just 2 factors of Default and non default. Finding shows that:

  • 13,158 loans are in outstanding with 7,260 defaulting or overdue loans and 5,898 maintaining good standing as normal loans. 36,841 were already fully paid. The pie charts displays more info on the distribution.
  • Distribution of Payment Plan among all loan status category showed no difference whereas their verification status are distinguished in the chart.
  • The Distribution of Loan Amounts and interest rate by Current Loan Status was plotted as an histogram. The Loan amount distribution showed a slightly right skewed distribution, showing lots of lower amounts and some higher amount in default. The interest rate distribution shows us an almost normal distribution and more rate in the middle defaulting.

• The Numeric Factors Correlation pane focused on numeric factors visual engaging Default/ loan overdue and more statistic. Finding shows that:

  • 36,149,044 loans volume principal are in outstanding with 2,054,452 overdue principal amount. 36 months shows the loan term of most defaulter.
  • The Portfolio at Risk report 6% volume for overdue >= 1 month and 55% customer counts for loan overdue >= 1 month which is a pointer that high number of clients are in overdue.
  • Correlation of Money and otherVariables with Default was displyed on the heat map to show their correlation with each other and the point biseral to show correlation with default, blue reflects very high correlation.

• The Categorical Factors Correlation pane focused on categorical factors visual engaging Default/ loan overdue. Finding shows that:

  • The crosstabs or stack bar charts shows the proportion of loans in overdue compared with normal loans and shows pattern in some of the factors sub-category distribution.
  • Crosstabs that has shown no difference is payment plan, the visual is a continum from Loan overview where Y payment plan was absent, this is because of its insufficiency in the dataset, and hence cannot be seen as a contributor to default.
  • Crosstabs that has shown patterns is the grade and subgrade factor, showing as grade rises so does default.
  • Others that may show some effect on default are Loan purpose, house ownership, employment length, verisfication status and loan term
  • While 100% overdue doesnt implies high impact, because sufficiency for those attribute may be low, hence the need for statistical test to determine their true impact

• The Statistical pane gives more information on the variables that have shown some influence on default. Finding shows that:

  • The Chisquare test measures the difference between the observed and expected frequencies of a categorical variable. Larger values indicate a greater discrepancy between the observed and expected distributions. The Logistic regression analysis was used to determine the best factors of high influence on default.
  • The Tukey HSD test results indicate that several numerical variables show significant differences between “Overdue” and “Normal” loans. This includes total accounts, open accounts, loan amount, outstanding principal, annual income, debt-to-income ratio, and revolving balance. The logistic regression will be used to emphasize what factor has high influence on default.
  • The logistic regression results indicate that several numerical variables, such as loan amount, outstanding principal, interest rate, annual income, debt-to-income ratio, revolving balance, and revolving utilization, significantly impact the likelihood of loan default.
  • Also, Higher loan grades and some sub-grades significantly increase the odds of default.Renting a home and certain loan terms or months of issuance are associated with higher or lower default odds. Verification status affects the likelihood of default, with verified loans being less risky.

• The Default Prediction pane focused on trend and geographical visual engaging Default/ loan overdue. Finding shows that:

  • The line chart showing default trend by loan issued date date and last payment date reveals that as loan were issued accross the years, default rises and visible straight line of probably a disbursement stop between 2011 and 2013. Conversely default lowers with last payment date while repayments continued till 2016, this is an expected nuance.
  • The forcast line plot of default count by their issued date has an Mean Absolute Error MAE of 14.92 this means that the typical prediction error (the average amount by which the forecasted default counts deviate from the actual counts) is about 14.92 defaults. While the RMSE i.e. Root Mean Squared Error indicates that the standard deviation of the prediction errors (residuals) is about 18.39 defaults .
  • The presence of an RMSE higher than the MAE but still relatively close in value suggests that while the model is generally accurate, there are instances where it can significantly deviate.
  • For decision-making or strategic planning regarding loan defaults, understanding the magnitude of these errors helps in assessing the level of confidence to be placed in the forecasts. An MAE and RMSE in the range provided might be acceptable depending on the total volume of defaults and the operational thresholds for risk management.
  • The geographical Distribution of Loan Default by state shows the proportion of default counts from different regions in the USA. The plot reviews that Nebraska with 55% has more than half of the defaulter settled there. Operations needs to find out if there was a recent economic issue or events that might have contributed to them defaulting their loans.