QuantModeling is a SaaS Tool for developing predictive or explanatory modeling, using the scoring concepts.

QuantModeling is an easy to use, low code software application for the development of predictive or explanatory machine learning models. The software leverages the power of the proven scorecard development methodology, using the Weight-of-Evidence concepts. Its in-built feature engineering engine allows for a rapid and easy visual representation of the relationship between the independent variables and the dependent variable. It can handle both binary and continuous dependent variables, and as such allows that the concepts of scoring can also be applied to a wider range of outcome predictions or use cases. Following the feature engineering, the software

allows the user to run an elastic net regression (binary outcome) or a linear regression (continuous outcome). For the binary prediction, QuantModeling will automatically produce industry-standard output such as a KPI/GINI coefficient, screening graphs and score-based performance tables. For the continuous dependent variable, an adapted regression is in-built, and the deviation rates are automatically provided.

QuantModeling provides the data scientist with a unique model development tool:

– It provides an end-to-end ML model development workflow, using the concept of Weight-of-Evidence and scoring. Its in-built scoring-based workflow, provides as output an easy-to-understand and easy-to-use scorecard.

– The software allows for a visual predictive or explanatory data pattern inspection, for the data scientists comprehension of the underlying data patterns, but also allows in-depth, graphically-supported discussions with subject matter experts – It holds a proprietary univariate feature analysis algorithm, which can automatically detect the underlying predictive/explanatory data patterns – The treatment of missing values and outliers is automatically taken care of (no need for imputation), or they can be specifically dealt with by the data scientist – The WoE can accommodate highly non-linear data patterns, should this be required – Both binary and continuous dependent variables can be accommodated for.

– The WoE feature transformation code, can be extracted from the software for direct use in other data science platforms such as R, Python or SAS.

– The in-built machine learning offers and immediate and powerful selection of the most important

predictors of the independent variables. A user can subsequently select or de-select variables, and re-run the regression.

– Provides automatically scoring-industry standard output such as KPI/GINI coefficient, a score-based performance tables and screening graphs.

– For the continuous dependent variable, the system provides the user with a visual and tabular overview of the deviation rate analysis, indicating the performance of the prediction.