DSC723 - Flight Delays Group Project v6

Data Understanding & EDA

Target Construction & Leakage Removal

Data Cleaning

Encoding

Data Splitting Method 1 — Holdout (70% Train / 30% Test)

Feature Selection Method A — Variance Threshold

Model 1 — Logistic Regression

Model 2 — Decision Tree

Model 3 — Random Forest

Splitting Method 2 — 10-Fold Stratified Cross-Validation for Model 2 — Decision Tree

Splitting Method 2 — 10-Fold Stratified Cross-Validation for Model 2 — Decision Tree

Data Splitting Method 1 — Holdout (70% Train / 30% Test)

Model 2 — Decision Tree

Feature Selection Method B — Chi-Square + Pearson Redundancy

Undersampling

Splitting Method 2 — 10-Fold CV for Model 1 (Logistic Regression)

Splitting Method 2 — 10-Fold CV for Model 3 (Random Forest)

Model 1 — Logistic Regression

Model 3 — Random Forest

Splitting Method 2 — 10-Fold CV for Model 1 (Logistic Regression)

Splitting Method 2 — 10-Fold CV for Model 3 (Random Forest)

RESULTS DASHBOARD — left: Holdout 70/30, right: 10-Fold CV. After executing, select all 6 Bar Charts (+ EDA charts) → right-click → Create Component → name it 'Flight Delay Dashboard'

KPI — Best model card (holdout, macro-F1): GroupBy → Sorter → top-1 → variables → Text View