Feature Selection & Model Reproducibility — Hands-on

Design robust QSAR/QSPR feature spaces and validation strategies that generalize. This module focuses on principled feature selection, collinearity control, leakage prevention, and reproducibility practices. You will implement end-to-end, audit-ready pipelines with nested cross-validation, proper reporting, and artifacts for reuse.

Feature Selection & Model Reproducibility for QSAR/QSPR

Help Desk · WhatsApp

Session Index

Session 1 — Data Hygiene & Collinearity Control Session 2 — Filter/Wrappers/Embedded & Dimensionality Reduction Session 3 — Validation: Nested CV, Leakage & Stability Session 4 — Mini Capstone: Reproducible FS Pipeline

Session 1

Fee: Rs 12800

Data Hygiene & Collinearity Control

Pre-FS checks & filters

missingness & imputation variance/near-zero variance scaling/normalization

Collinearity diagnostics

correlation matrices VIF thresholds cluster-based pruning

Train/test hygiene

scaffold/stratified splits no peeking policies pipelines for transformations

Session 2

Fee: Rs 15800

Filter/Wrappers/Embedded & Dimensionality Reduction

Filter methods

univariate tests mutual information mRMR

Wrappers & embedded

RFE/RFECV Boruta/SHAP Lasso/Elastic Net

Dimension reduction

PCA/PLS UMAP for EDA targets & scaling effects

Session 3

Fee: Rs 18800

Validation: Nested CV, Leakage & Stability

Leakage-safe pipelines

Pipeline/ColumnTransformer fit/transform boundaries temporal/scaffold splits

Model selection the right way

nested cross-validation grouped CV bootstrap stability

Sanity checks & AD

y-scrambling permutation tests applicability domain

Session 4

Fee: Rs 22800

Mini Capstone: Reproducible FS Pipeline

Build: filter → FS → model → validate

Theory + Practical

Report & artifacts

config YAML/JSON random seeds & env lock model/FS cards

Deliverables

reproducible notebook/script metrics + stability plots pipeline diagram

PDF

Feature Selection & Model Reproducibility for QSAR/QSPR | Leakage-Safe, Nested-CV Pipelines

Feature Selection & Model Reproducibility — Hands-on