Future Plan

Future Plan

Purpose

This page enumerates candidate enhancements and methodological extensions that will deepen triangulation capabilities and analytical rigor within CRISP-T.

Planned / Proposed Methods

Area	Enhancement	Rationale for Triangulation
Ensemble Models	Random Forests, Gradient Boosting (XGBoost integration already optional)	Stabilize variable importance; cross-check decision tree rules
Regularized Regression	Lasso / Ridge / Elastic Net	Handle high-dimensional text-derived numeric features, support selective coding
Mixed-Effects Models	Hierarchical regression (e.g., patient within clinic)	Model nested qualitative contexts
Advanced Clustering	DBSCAN, Hierarchical clustering	Detect non-spherical clusters complementing K-Means and TDABM
Topic Modeling	Neural topic models / BERTopic	Capture contextual embedding-based themes
Abstractive Summarization	Transformer summarizers	Thematic condensation for memo drafting (human-validated)
Sentiment Dimensions	Emotion / stance analysis	Richer qualitative affect coding
Causal Inference	Propensity score / causal forests	Test robustness of emergent explanatory relations
Interactive Visualization	Web dashboard combining TDABM + topic heatmaps	Real-time multi-modal exploration
Rule Export	JSON path export from decision trees	Programmatic comparison with qualitative category networks
Radius Optimization	Automated TDABM radius tuning	Reduce analyst bias in structural inference
Memo Automation	Template generation populated from metadata	Accelerate audit trail construction
Counterfactual Search	Generate alternative semantic queries	Mitigate confirmation bias

Prioritization Criteria

1. Contribution to methodological rigor (grounded theory alignment).
2. Enhancement of cross-modal interpretability.
3. Low risk of misinterpretation or over-automation.
4. Community demand (issues, stars, citations).

Theoretical Integration Roadmap

Phase	Focus	Outcome
Short-term	Ensemble + regularized models	Stable confirmatory signals
Mid-term	Mixed-effects + advanced clustering	Context-sensitive saturation
Long-term	Causal layers + interactive dashboards	Theory refinement & dissemination

How to Contribute

• Open an issue describing the proposed method and its triangulation use-case.
• Provide minimal reproducible examples and expected metadata outputs.
• Reference this plan and justify priority alignment.

References

• Mettler et al. (2025) — Methodological reflection guiding computational augmentation.
• Rudkin & Dlotko (2024) — Structural exploration, motivating shape-aware enhancements.

See Also

• Predictive Methods
• DecisionTrees
• Regression
• Topological Data Analysis
• Human Sense-making and triangulation