Task: Isolation Forest model on Shuttle dataset

forest.py

@@ -0,0 +1,74 @@
+import pandas as pd
+from scipy.io import arff
+import matplotlib.pyplot as plt
+from sklearn.preprocessing import RobustScaler
+from sklearn.ensemble import IsolationForest
+from sklearn.metrics import precision_score, recall_score, f1_score, roc_auc_score, ConfusionMatrixDisplay
+
+#Loading the shuttle dataset
+data, meta = arff.loadarff("shuttle.arff")
+df = pd.DataFrame(data)
+target = pd.Series(df['class']).astype(int)
+
+#Class 1 and 4 make up 93.94% of observations, hence they are classified as normal or 1
+anomalies = target.apply(lambda x: 1 if x == 1 or x == 4 else -1)
+
+#Dropping the class column as it is not needed for the model
+df.drop('class', axis = 'columns', inplace = True)
+
+
+#Scaling the data using RobustScaler
+scale = RobustScaler().set_output(transform = "pandas")
+df_scaled = scale.fit_transform(df)
+
+#Calculating actual contamination:
+total_entries = len(target)
+no_of_anomalies = (anomalies == -1).sum()
+contamination = no_of_anomalies/total_entries
+
+
+#Applying isolation forest model, using approximate contamination of 6.05%
+iso_forest = IsolationForest(n_estimators = 100, contamination = 0.0605, max_samples = 256, random_state = 42)
+iso_forest.fit(df_scaled)
+data = df.loc[df.index].copy()   #data stores the input data along with corresponding predicted labels
+data['anomaly_score'] = iso_forest.decision_function(df_scaled)
+data['anomaly'] = iso_forest.predict(df_scaled)
+
+
+print("Here are the number of normal and anomaly values:")
+print(data['anomaly'].value_counts())
+
+#Now calculating precision, recall, f1 score, ROC-AUC value
+precision = precision_score(anomalies, data['anomaly'], pos_label = -1)
+print("Precision: ", precision)
+
+recall = recall_score(anomalies, data['anomaly'], pos_label = -1)
+print("Recall: ", recall)
+
+f1 = f1_score(anomalies, data['anomaly'], pos_label = -1)
+print("F1 Score: ", f1)
+
+roc_auc = roc_auc_score(anomalies, data['anomaly_score'])
+print("ROC-AUC value: ", roc_auc)
+
+
+#Plotting a Confusion Matrix
+ConfusionMatrixDisplay.from_predictions(anomalies, data['anomaly'], display_labels=["Anomaly", "Normal"], cmap=plt.cm.Blues)
+
+
+# Visualization of the results through a scatter plot
+
+plt.figure(figsize=(15, 10))
+
+#'normal' stores all instances that are classified as (1)
+normal = data[data['anomaly'] == 1]
+plt.scatter(normal.index, normal['anomaly_score'], label='Normal', alpha=0.2)
+
+#'anomalies' stores all instances that are classified as (-1)
+anomalies = data[data['anomaly'] == -1]
+
+plt.scatter(anomalies.index, anomalies['anomaly_score'], label='Anomaly', alpha=0.3)
+plt.xlabel("Instance")
+plt.ylabel("Anomaly Score")
+plt.legend()
+plt.show()