IntelScoring - AI Development of Scoring Models

IntelScoring offers powerful data pre-processing tools that automatically or manually handle missing values to ensure high-quality input for model development. The system can apply statistical techniques like mean, median, or predictive imputation, or allow analysts to manually fill in gaps based on business rules. Custom logic and formulas can be used for advanced cases, ensuring adaptability across industries. All replacements are logged for transparency and compliance. Continuous data quality checks help identify anomalies early, improving model accuracy and stability. This ensures reliable, audit-ready data preparation that supports fast and effective scoring model creation.

IntelScoring enables the creation of new variables by applying custom formulas to existing borrower characteristics, enhancing model accuracy and flexibility. Analysts can combine multiple input fields—such as income, age, or credit history—into derived indicators that better capture risk factors. This supports the development of more predictive features tailored to specific business needs. The system allows both manual input and automated generation of calculated fields. Each new variable is fully traceable, ensuring transparency and regulatory compliance. This functionality empowers users to build richer, more meaningful datasets for robust scoring models.

IntelScoring provides flexible tools for creating statistical samples used in model training and testing, enabling faster and more efficient model development. Users can define time periods, select subsets of data, and adjust sample sizes to match specific modeling goals. The system supports balancing the ratio of “good” and “bad” cases to improve classification performance. Sampling parameters can be saved and reused, ensuring consistency across model iterations. This streamlined approach helps reduce data preparation time while maintaining high-quality input for accurate, stable scoring models.

IntelScoring enables manual editing of variable category ranges after automatic binning, providing greater control over model inputs. As categories are adjusted, the system automatically recalculates Weight of Evidence (WoE) values to maintain statistical validity. Visual graphs display WoE values alongside distributions of “good” and “bad” borrowers, helping analysts make informed adjustments. This feature supports fine-tuning of scoring models based on expert judgment or regulatory needs. It enhances transparency and ensures that category groupings align with real-world risk patterns.

IntelScoring allows users to create combined variables by merging two or more borrower characteristics, such as marital status and employment type. This enhances the model's ability to capture complex patterns and interactions that single variables may miss. Combined variables can be formed from both categorical and numerical data, offering broad flexibility. The system supports both manual and automated combinations, depending on analytical goals. These enriched features often lead to improved model performance and deeper insights into borrower behavior. All combined variables are fully documented for transparency and reproducibility.

IntelScoring allows analysts to manually select, include, or exclude specific factors before initiating automated variable selection. This ensures that domain knowledge or regulatory requirements can guide the modeling process. Users can lock in key variables they consider critical or remove those deemed irrelevant or biased. The system then performs automated selection on the remaining pool of factors. This hybrid approach balances expert input with algorithmic efficiency, leading to more accurate and compliant scoring models.

IntelScoring provides a comprehensive set of tools to evaluate model performance, including ROC curve visualization, Gini coefficient, AUC, determination coefficients, and Kolmogorov-Smirnov statistics. These metrics help assess the model’s ability to distinguish between reliable and risky borrowers. Results are presented for both training and testing datasets, ensuring stability and generalizability. Visual outputs make it easy to interpret model quality at a glance. This analytical depth supports transparent, data-driven decisions and regulatory compliance.

IntelScoring enables users to modify the set of factors in a scoring model even after it has been built, offering flexibility for refinement and improvement. Once adjustments are made, the system automatically recalculates all model parameters and effectiveness metrics. This allows analysts to test different configurations and optimize performance without rebuilding models from scratch. It also supports regulatory or business-driven changes that require rapid updates. The process is fully transparent and helps ensure that models remain accurate and aligned with evolving requirements.