Task Isolation Forest Model on ODDS Shuttle dataset

[NikhByte]

Pull Request

Related Issue

Closes #16

---

Task Summary

Provide a brief overview of your implementation.

• What did you implement?
• I trained a model on the shuttle ODDS dataset using the isolation forest to detect the the anomalies in the dataset. In this i learned the benchmarking and the evaluation of the anomaly detection in the unsupervised anomaly detection algorithm isolation forest.
• What approach did you follow?
• Firstly after importing the data and converted the data into the DataFrame i looked for the missing valeus(NaN) luckily there weren't any so there was almost no preprocessing required.
  Major Challenge i faced during training of the model was recalling the correct syntax and some libs, I took some help from the medium page.To decide the best parameters values like n_estimators,contamination and sample size i did some hard yakka, by running the loops to decide the n_estimators in range(1,700,50) luckily i got the highest f1 scores for values 101 and 151 but i took 100 because it had higher f1_score as well as auc_value both tending to one if i took approximaiton upto 2 decimal places.

---

Dataset

• Mammography
• Shuttle

Dataset Source:

---

Preprocessing

Describe any preprocessing performed.

• Missing value handling - No,It has nil NaN values confirmed it by df.isnull().sum()==0
• Feature scaling - No isolation forest can handle without scaling.
• Encoding - No data contained the numeric values so not needed.
• Feature selection - No verly low dimensionality.

---

Model Configuration

List the important hyperparameters used.

Hyperparameter	Value
n_estimators	100
contamination	0.075
max_samples	256
max_features	1
random_state	42

---

Evaluation Results

Metric	Value
Precision	0.9332
Recall	0.9789
F1-score	0.955
ROC-AUC (Optional)	0.9975

---

Visualizations

Attach at least 2 plots from your analysis.

---

Key Observations

Briefly summarize:

• What worked well?
• Preprocessing,luckily it needed no preprocessing it featured low dimensional features(only 9) and all the data was numeric and there wasn't any NaN values.
• Which hyperparameter had the biggest impact?
• Contamination value.increasing it slightly helped me achieved the good f1_score,recall slight more auc_val.
• Challenges faced (if any)
• Since new to Isolation forest faced slight issues with syntax and libs.

---

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