About

This canvas is a comprehensive data science workflow focused on prospect evaluation and machine learning model diagnostics. It starts by loading and preparing prospect data, including encoding features and splitting into training and test sets. Several classification models are trained and optimized, including Random Forest, Logistic Regression, XGBoost, and LightGBM, often using cross-validation and hyperparameter tuning for robust performance. The canvas includes detailed model evaluation steps such as computing accuracy, ROC AUC, precision-recall curves, calibration curves, Brier scores, confusion matrices, and residual analysis. It further investigates feature importance and stability across cross-validation splits to understand key predictors. Subgroup analysis is performed by categorical segments (e.g., regions) to assess how model performance varies across different groups. Visualization blocks produce histograms, correlation heatmaps, violin plots, and various plots for model diagnostics. Finally, the canvas includes domain-specific simulation and data storage in Snowflake, integrating data science with business domain knowledge. Overall, it is a modular, end-to-end pipeline combining data preparation, model training, performance evaluation, interpretability, subgroup assessment, and domain simulation for prospect evaluation.