Issue #17

PULL_REQUEST_TEMPLATE.md

@@ -8,90 +8,78 @@ Closes #
 
 ## Task Summary
 
-Provide a brief overview of your implementation.
-
-- What did you implement?
-- What approach did you follow?
+- **What did you implement?** Implemented an unsupervised anomaly detection system using the Isolation Forest algorithm to flag anomalous shuttle flights.
+- **What approach did you follow?** Followed an experimental approach: established a baseline with default parameters, analyzed the see-saw trade-off between Precision and Recall by adjusting strictness parameters, tested feature scaling, and ultimately executed an automated Grid Search to locate the optimal safety configuration.
 
 ---
 
 ## Dataset
 
 - [ ] Mammography
-- [ ] Shuttle
+- [x] Shuttle
 
 Dataset Source:
 
 ---
 
 ## Preprocessing
 
-Describe any preprocessing performed.
-
-Examples:
-- Missing value handling
-- Feature scaling
-- Encoding
-- Feature selection
+- There were no missing values in the dataset.
+- **Feature Scaling Evaluation:** We integrated data standardization using `StandardScaler` to evaluate the model's sensitivity to feature magnitudes. The experiments successfully validated that the Isolation Forest algorithm is inherently scale-invariant. Because the model relies on recursive, axis-aligned isolation trees rather than geometric distance metrics, scaling preserves the exact relative separation paths of the anomalies. This is an exceptional characteristic for our pipeline, as it proves the model achieves peak predictive performance with reduced preprocessing overhead.
 
 ---
 
 ## Model Configuration
 
-List the important hyperparameters used.
+*(For best "Safety-First" model)*
 
 | Hyperparameter | Value |
 |---------------|-------|
-| n_estimators | |
-| contamination | |
-| max_samples | |
-| max_features | |
-| random_state | |
+| n_estimators | 100 |
+| contamination | 0.50 |
+| max_samples | 256 |
+| max_features | 1.0 |
+| random_state | 42 |
 
 ---
 
 ## Evaluation Results
 
 | Metric | Value |
 |--------|-------|
-| Precision | |
-| Recall | |
-| F1-score | |
-| ROC-AUC (Optional) | |
+| Precision | 0.40 |
+| Recall | 0.94 |
+| F1-score | 0.56 |
+
 
 ---
 
 ## Visualizations
 
-Attach **at least 2 plots** from your analysis.
+**Confusion Matrix — Best Safety Model**
 
-Examples:
-- PCA visualization
-- Anomaly score distribution
-- Confusion Matrix
-- Correlation heatmap
-- Feature distributions
-- Hyperparameter comparison
-- Precision/Recall/F1 comparison
+![Confusion Matrix](./confusion_matrix.png)
+
+**Impact of Contamination Threshold on Anomaly Recall**
+
+![Hyperparameter Impact](./hyperparameter_impact.png)
 
 ---
 
 ## Key Observations
 
-Briefly summarize:
-
-- What worked well?
-- Which hyperparameter had the biggest impact?
-- Any interesting findings?
-- Challenges faced (if any)
+- **What worked well?** Increasing the `contamination` parameter significantly expanded the classification envelope, maximizing our Recall to 0.94 (catching 94% of shuttle system anomalies).
+- **Which hyperparameter had the biggest impact?** `contamination` had the absolute biggest impact on shifting the see-saw balance between precision and recall.
+- **Any interesting findings?** Increasing `n_estimators` beyond 200–300 resulted in diminishing returns and minor score degradation due to algorithmic plateauing. Furthermore, setting `contamination` too high combined with tiny sample sizes caused "swamping," where normal data points overwhelmed the trees' ability to isolate actual anomalies.
+- **Challenges faced:** Overcoming short-term memory clears when switching kernel environments in VS Code, requiring structured notebook tracking.
 
 ---
 
 ## Checklist
 
-- [ ] Code runs successfully
-- [ ] Notebook (`.ipynb`) included
-- [ ] Code is well-commented
-- [ ] README/documentation updated
-- [ ] At least **2 plots** included
-- [ ] PR is linked to the corresponding issue
+- [x] Code runs successfully
+- [x] Notebook (`.ipynb`) included
+- [x] Code is well-commented
+- [x] README/documentation updated
+- [x] At least **2 plots** included
+- [x] PR is linked to the corresponding issue

README.md

@@ -1,15 +1,118 @@
-# CogniOS – Tasks
+DIFFERENT CASES USED 
+1-
+iso_forest = IsolationForest(
+contamination =0.4,
+max_samples = 256, 
+random_state=42,
+max_features = 1.0 , 
+n_estimators = 120)
 
-This branch contains tasks designed to help contributors learn the concepts and technologies used in CogniOS before contributing to the main project.
+Classification Report:
+              precision    recall  f1-score   support
 
-## Workflow
+          -1       0.45      0.84      0.59     12414
+           1       0.94      0.72      0.82     45586
 
-1. Check your assigned GitHub issue.
-2. Create a new branch from `Tasks`.
-3. Complete the assigned task.
-4. Open a Pull Request **to the `Tasks` branch**.
-5. Mention `Closes #<issue_number>` in your PR description.
+2-
+iso_forest = IsolationForest(
+contamination =0.4,
+max_samples = 400, 
+random_state=42,
+max_features = 1.0 , 
+n_estimators = 200)
 
-Please follow the repository's Pull Request template while submitting your solution.
+Classification Report:
+              precision    recall  f1-score   support
 
-Happy learning! 
+          -1       0.43      0.81      0.56     12414
+           1       0.93      0.71      0.81     45586
+
+
+3-(using standard values)
+iso_forest = IsolationForest(
+contamination =0.4,
+max_samples = 256, 
+random_state=42,
+max_features = 1.0 , 
+n_estimators = 100)
+
+Classification Report:
+              precision    recall  f1-score   support
+
+          -1       0.46      0.86      0.60     12414
+           1       0.95      0.73      0.82     45586
+
+
+ 4-(assuming most amount of contamination)
+ (most recall obtained)
+ iso_forest = IsolationForest(
+contamination =0.5,
+max_samples = 256, 
+random_state=42,
+max_features = 1.0 , 
+n_estimators = 100)          
+
+Classification Report:
+              precision    recall  f1-score   support
+
+          -1       0.40      0.94      0.56     12414
+           1       0.97      0.62      0.76     45586
+
+5 - (actual amount of contamination)
+iso_forest = IsolationForest(
+contamination =0.21,
+max_samples = 256, 
+random_state=42,
+max_features = 1.0 , 
+n_estimators = 100)
+
+Classification Report:
+              precision    recall  f1-score   support
+
+          -1       0.54      0.53      0.54     12414
+           1       0.87      0.88      0.88     45586
+
+
+6 - (inc no of trres anfd reducing samples )
+
+iso_forest = IsolationForest(
+contamination =0.5,
+max_samples = 100, 
+random_state=42,
+max_features = 1.0 , 
+n_estimators = 400)
+
+Classification Report:
+              precision    recall  f1-score   support
+
+          -1       0.37      0.86      0.52     12414
+           1       0.94      0.60      0.73     45586
+
+7- 
+
+iso_forest = IsolationForest(
+contamination =0.5,
+max_samples = 50, 
+random_state=42,
+max_features = 1.0 , 
+n_estimators = 290) 
+
+ precision    recall  f1-score   support
+
+          -1       0.38      0.89      0.53     12414
+           1       0.95      0.61      0.74     45586
+
+
+8 --
+contamination =0.5,
+max_samples = 50, 
+random_state=42,
+max_features = 1.0 , 
+n_estimators = 400)
+
+Classification Report:
+              precision    recall  f1-score   support
+
+          -1       0.38      0.88      0.53     12414
+           1       0.95      0.60      0.74     45586
+