Minitab Salford Predictive Modeler
The Salford Predictive Modeler software suite offers a range of powerful tools and features that are unmatched by any other software. It includes engines such as CART®, MARS®, TreeNet®, and Random Forests®, along with advanced automation and modeling capabilities.
The suite's data mining technologies cover various areas including classification, regression, survival analysis, missing value analysis, data binning, and clustering/segmentation. SPM algorithms are highly regarded in the field of data science, being recognized as indispensable by experts. One of the key advantages of the SPM software suite is its automation, which significantly speeds up the model building process. It automates substantial parts of model exploration and refinement, making it easier for analysts to review and compare results from different modeling strategies.
Features of Minitab Salford Predictive Modeler
The software offers a range of modeling engines, including CART decision trees, TreeNet gradient boosting, Random Forests tree ensemble, MARS nonlinear regression splines, and GPS regularized regression. It also includes features like RuleLearner for combining accuracy and interpretability, ISLE model compression, and over 70 automation routines for model building and experimentation. Specific features for each modeling engine are provided, such as hotspot identification and variable importance measures for CART, graphical understanding and variable assessment for MARS, partial dependency plots and regression loss functions for TreeNet, and classification capabilities and outlier detection for Random Forests. The software also provides automation tools for tasks like feature selection, cross-validation, and exploring different learning and testing partitions.
| General Features | Modeling Engine: CART® decision trees | |
| Modeling Engine: TreeNet® gradient boosting | ||
| Modeling Engine: Random Forests® tree ensemble | ||
| Modeling Engine: MARS® nonlinear regression splines | ||
| Modeling Engine: GPS regularized regression (LASSO, Elastic Net, Ridge, etc.) | ||
| Modeling Engine: RuleLearner®, combining TreeNet's accuracy with regression interpretability | ||
| Modeling Engine: ISLE model compression | ||
| Over 70 pre-packaged automation routines for advanced model building and experimentation | ||
| Tools to alleviate repetitive tasks, enabling analysts to focus on the creative aspects of model development | ||
| Support for Open Minitab Worksheet (.MTW) functionality | ||
| CART® Features | Identify hotspots for detecting the most crucial sections of the tree and their corresponding tree rules | |
| Determine variable importance measures to comprehend the significance of variables within the tree | ||
| Deploy the model for real-time or other prediction generation purposes | ||
| Implement user-defined splits at any node in the tree | ||
| Utilize differential lift modeling to evaluate the effectiveness of a treatment or intervention. | ||
| Automation tools for model tuning and other experiments including | Employ automatic recursive feature elimination for advanced variable selection | |
| Explore different prior probabilities to enhance accuracy rates for the more significant class | ||
| Conduct repeated cross-validation for robust model evaluation | ||
| Construct CART models on bootstrap samples for improved model performance | ||
| Develop two linked models: one predicting a binary event and the other predicting a numeric value (e.g., purchase likelihood and spending amount) | ||
| Evaluate the effects of different learning and testing partitions on model performance | ||
| MARS® Features | Gain a graphical understanding of how variables impact the model response | |
| Assess the importance of individual variables or groups of interacting variables | ||
| Deploy the model for real-time prediction generation or other purposes | ||
| Automation tools for model tuning and other experiments including | Utilize automatic recursive feature elimination for advanced variable selection | |
| Automatically evaluate the impact of incorporating interactions in the model | ||
| Effortlessly identify the optimal minimum span value | ||
| Conduct repeated cross-validation | ||
| Explore the effects of various learning and testing partitions | ||
| TreeNet® Features | Graphically understand variable impact on model response through partial dependency plots | |
| Regression loss functions: least squares, least absolute deviation, quantile, Huber-M, Cox survival, Gamma, Negative Binomial, Poisson, Tweedie | ||
| Classification loss functions: binary or multinomial | ||
| Differential lift modeling (also known as "uplift" or "incremental response") | ||
| Column subsampling for improved model performance and faster runtime | ||
| Regularized Gradient Boosting (RGBOOST) for increased accuracy | ||
| RuleLearner: Create interpretable regression models by combining TreeNet gradient boosting and regularized regression (LASSO, Elastic Net, Ridge, etc.) | ||
| ISLE: Build smaller and more efficient gradient boosting models using regularized regression (LASSO, Elastic Net, Ridge, etc.) | ||
| Variable Interaction Discovery Control | Determine definitively whether or not interactions of any degree need to be included | |
| Control the interactions allowed or disallowed in the model with Minitab’s patented interaction control language | ||
| Discover the most important interactions in the model | ||
| Calibration tools for rare-event modeling | ||
| Automation tools for model tuning and other experiments including | Automatic recursive feature elimination for advanced variable selection | |
| Automated experimentation with different learning rates | ||
| Control the level of interactions within the model | ||
| Build linked models: one predicting a binary event and the other predicting a numeric value (e.g., predicting purchase likelihood and spending amount) | ||
| Determine the optimal parameters for regularized gradient boosting models | ||
| Perform a stochastic search for the core gradient boosting parameters | ||
| Explore the effects of different learning and testing partitions | ||
| Random Forests® Features | Classification, regression, and clustering capabilities | |
| Outlier detection functionality | ||
| Graphical tools like proximity heat maps and multi-dimensional scaling for identifying clusters in classification problems | ||
| Parallel Coordinates Plot for visualizing predictor values and class assignments | ||
| Advanced techniques for handling missing values | ||
| Unsupervised learning using Random Forest and hierarchical clustering | ||
| Variable importance measures to determine significant variables in the model | ||
| Real-time deployment and prediction generation capabilities | ||
| Automation tools for model tuning and other experiments including | Automatic recursive feature elimination for advanced variable selection | |
| Flexible adjustment of random subset size at each split in each tree | ||
| Analysis of the impact of different bootstrap sample sizes | ||
| Evaluation of the impact of different learning and testing partitions |