devlup-labs · abhishekreddy22 · Jul 10, 2026 · Jul 16, 2026 · Jul 16, 2026 · Jul 16, 2026
diff --git a/PULL_REQUEST_TEMPLATE.md b/PULL_REQUEST_TEMPLATE.md
@@ -2,7 +2,7 @@
 
 ## Related Issue
 
-Closes # 
+Closes # 12 
 
 ---
 
@@ -14,13 +14,15 @@ Provide a brief overview of your implementation.
 Implemented the Isolation Forest model for detecting anomalous mammography data
 
 - What approach did you follow?
+
 I extracted the data from a csv file,
 selected the features to detect the anomaly
 used dropna() to eliminate rows with missing values
 used StandardScaler to scale features
 split the data into test and train (test=0.1)
 trained the isolation forest model
 used iso_forest.predict() to get the number of anomaly
+used matplotlib to plot the data
 
 ---
 
@@ -35,20 +37,12 @@ Dataset Source: OpenML
 
 ## Preprocessing
 
-Describe any preprocessing performed.
-
 Handled missing values with pandas' built in function df.dropna()
 which drops rows with NaN
 
-Selected only first 6 rows for isolation forest model as
-originally the dataset was created for classification and the 7th column contained labels
+Found out using SHAP, that col 4/5/6 were the major contributors to the model and rest of the columns were just noise, omitting col 1/2/3 increased f1_score drastically
 
-Used Standard Scaler for scaling the data for better accuracy
-Examples:
-- Missing value handling
-- Feature scaling
-- Encoding
-- Feature selection
+Omitted using Standard Scaler as the the data was already scaled
 
 ---
 
@@ -59,9 +53,9 @@ List the important hyperparameters used.
 | Hyperparameter | Value |
 |---------------|-------|
 | n_estimators | 100 |
-| contamination | 0.009 |
+| contamination | 0.023 |
 | max_samples | 256 |
-| max_features | 6 |
+| max_features | 3 |
 | random_state | 42 |
 
 ---
@@ -70,37 +64,43 @@ List the important hyperparameters used.
 
 | Metric | Value |
 |--------|-------|
-| Precision | |
-| Recall | |
-| F1-score | |
+| Precision |0.615|
+| Recall |0.473|
+| F1-score |0.535|
 | ROC-AUC (Optional) | |
 
 ---
 
 ## Visualizations
 
 Attach **at least 2 plots** from your analysis.
-![alt text](image.png)
-Examples:
-- PCA visualization
-- Anomaly score distribution
-- Confusion Matrix
-- Correlation heatmap
-- Feature distributions
-- Hyperparameter comparison
-- Precision/Recall/F1 comparison
+---
+**Confusion Matrix**
+![alt text](confusion_matrix.png)
+---
+**Scatter Plot**
+![alt text](anomaly_scatter_plot.png)
+---
 
 ---
 
 ## Key Observations
 
 Briefly summarize:
 
-- What worked well?
-- Which hyperparameter had the biggest impact?
-Contamination had the biggest impact, as it directly affects the number of anomalies detected
-- Any interesting findings?
-- Challenges faced (if any)
+- What worked well?\
+The most important function that helped tune the model better was precision_recall_curve(), which helped me know the exact threshold value, instead of blindly going for predict()
+
+- Which hyperparameter had the biggest impact?\
+3 Main hyperparameters **Contamination|n_estimators|max_samples** had high impacts on f1score, especially **Contamination** and **max_features**
+- Any interesting findings?\
+
+- Challenges faced (if any)\
+Most of the challenge faced was getting familiar with the syntax of different libraries, and researching how to improve the f1_score took a lot of time
+initial f1_score: 0.264 
+final f1_score: 0.535
+
+Understanding how to evaluate the model (i.e implementing the precision_score etc) was a big challenge due to different conventions among OpenML|Isolation_Forest|Precision_Score
 
 ---
 
@@ -109,6 +109,6 @@ Contamination had the biggest impact, as it directly affects the number of anoma
 - [x] Code runs successfully
 - [ ] Notebook (`.ipynb`) included
 - [x] Code is well-commented
-- [ ] README/documentation updated
-- [ ] At least **2 plots** included
-- [ ] PR is linked to the corresponding issue
+- [x] README/documentation updated
+- [x] At least **2 plots** included
+- [x] PR is linked to the corresponding issue
diff --git a/image-1.png → anomaly_scatter_plot.png b/image-1.png → anomaly_scatter_plot.png
diff --git a/image.png → confusion_matrix.png b/image.png → confusion_matrix.png
diff --git a/i_forest.py b/i_forest.py
@@ -1,4 +1,5 @@
 # new things learnt
+# got familiarized with syntax of new libraries and standard practices
 # precision_score(y_true, y_pred) both these values must be same for binary -> same tuple
 import pandas as pd
 import numpy as np

diff --git a/mammography.csv b/mammography.csv
@@ -1,4 +1,4 @@
-0.23001961,5.0725783,-0.27606055,0.83244412,-0.37786573,0.4803223,'-1'
+0.23001961,5.0725783,-0.27606055,0.83244412,-0.37786573,0.4803224,'-1'
 0.15549112,-0.16939038,0.67065219,-0.85955255,-0.37786573,-0.94572324,'-1'
 -0.78441482,-0.44365372,5.6747053,-0.85955255,-0.37786573,-0.94572324,'-1'
 0.54608818,0.13141457,-0.45638679,-0.85955255,-0.37786573,-0.94572324,'-1'

diff --git a/ni.py b/ni.py
diff --git a/pyproject.toml b/pyproject.toml
@@ -6,5 +6,6 @@ readme = "README.md"
 requires-python = ">=3.13"
 dependencies = [
     "matplotlib>=3.11.0",
+    "scikit-learn>=1.9.0",
     "seaborn>=0.13.2",
 ]