Kaggle Machine Learning Competition: Phishing Email Detection#
Link to private Kaggle competition
Import modules#
import os
import numpy as np
import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
import matplotlib_inline.backend_inline
matplotlib_inline.backend_inline.set_matplotlib_formats("svg")
plt.rcParams["figure.facecolor"] = "white"
Read and split the data#
from sklearn.model_selection import train_test_split
def process_data(data):
# return data.str.encode("ascii", "ignore").str.decode("ascii")
return data
home_dir = os.path.expanduser("~")
local_dir = f"{home_dir}/github/KaggleCompetitionSpamNotSpam/datasets"
train_data = pd.read_csv(f"{local_dir}/train.csv", index_col=0)
test_data = pd.read_csv(f"{local_dir}/test.csv", index_col=0)
train_data["email"] = process_data(train_data["email"])
test_data["email"] = process_data(test_data["email"])
train, val = train_test_split(train_data, test_size=0.15, random_state=17, stratify=train_data["label"])
print(f"Train data shape: {train.shape}")
print(f"Validation data shape: {val.shape}")
Train data shape: (6386, 4)
Validation data shape: (1127, 4)Feature extraction - Vectorize words#
from sklearn.feature_extraction.text import CountVectorizer
count_vect = CountVectorizer(analyzer="word", ngram_range=(1, 2), lowercase=True)
X_train_counts = count_vect.fit_transform(train["email"])
ser_features_names = pd.Series(count_vect.get_feature_names_out())
ser_features_names.to_frame("feature_names").info()
class 'pandas.core.frame.DataFrame'
RangeIndex: 663546 entries, 0 to 663545
Data columns (total 1 columns):
# Column Non-Null Count Dtype
--- ------ -------------- -----
0 feature_names 663546 non-null object
dtypes: object(1)
memory usage: 5.1+ MBFeature extraction - Find importance of words#
from sklearn.feature_extraction.text import TfidfTransformer
tfidf_transformer = TfidfTransformer()
X_train_tfidf = tfidf_transformer.fit_transform(X_train_counts)
ser_features_tfidf = pd.Series(tfidf_transformer.idf_)
ser_features_tfidf.to_frame("importance").info()
class 'pandas.core.frame.DataFrame'
RangeIndex: 663546 entries, 0 to 663545
Data columns (total 1 columns):
# Column Non-Null Count Dtype
--- ------ -------------- -----
0 importance 663546 non-null float64
dtypes: float64(1)
memory usage: 5.1 MBModel Selection, training, and testing#
from sklearn.linear_model import SGDClassifier
from sklearn import metrics
text_clf = SGDClassifier(loss="hinge", penalty="l2",
alpha=1e-6, random_state=17,
max_iter=2e3, tol=1e-12,
fit_intercept=True)
text_clf.fit(X_train_tfidf, train["label"])
y_pred = text_clf.predict(tfidf_transformer.transform(count_vect.transform(val["email"])))
print(metrics.classification_report(val["label"], y_pred, target_names=["not spam", "spam"], digits=4))
precision recall f1-score support
not spam 0.9952 0.9952 0.9952 838
spam 0.9862 0.9862 0.9862 289
accuracy 0.9929 1127
macro avg 0.9907 0.9907 0.9907 1127
weighted avg 0.9929 0.9929 0.9929 1127Visualize the results#
fig, axes = plt.subplots(nrows=1, ncols=2, figsize=(10, 4))
cm = metrics.confusion_matrix(val["label"], y_pred)
cm_display = metrics.ConfusionMatrixDisplay(cm).plot(ax=axes[0], cmap=plt.cm.Oranges)
precision, recall, _ = metrics.precision_recall_curve(val["label"], y_pred)
pr_display = metrics.PrecisionRecallDisplay(precision=precision, recall=recall).plot(ax=axes[1], color="coral")
Confusion Matrix and Precision vs Recall Plot
Submit the results#
y_submit = text_clf.predict(tfidf_transformer.transform(count_vect.transform(test_data["email"])))
df_submit = test_data.reset_index()[["id"]].copy().set_index("id")
df_submit["label"] = y_submit
df_submit.to_csv(f"{local_dir}/submit_REV2.csv")