How to build a Spam Classifier in python and sklearn
![Author](/_next/image?url=%2Favatar.jpg&w=48&q=75)
Milind Soorya / September 13, 2021
9 min read
![Build a Spam Classifier in python banner](/_next/image?url=%2F_next%2Fstatic%2Fmedia%2Fbanner.5a432c2e.jpg&w=3840&q=75)
Introduction
The upsurge in the volume of unwanted emails called spam has created an intense need for the development of more dependable and robust antispam filters. Any promotional messages or advertisements that end up in our inbox can be categorised as spam as they don't provide any value and often irritates us.
Overview of the Dataset used
We will make use of the SMS spam classification data.
The SMS Spam Collection is a set of SMS tagged messages that have been collected for SMS Spam research. It contains one set of SMS messages in English of 5,574 messages, tagged according to being ham (legitimate) or spam.
The data was obtained from UCI’s Machine Learning Repository, alternatively, I have also uploaded the dataset and completed Jupiter notebook onto my GitHub repo.
In this article, we'll discuss:
Data processing
- Import the required packages
- Loading the Dataset
- Remove the unwanted data columns
- Preprocessing and Exploring the Dataset
- Build word cloud to see which message is spam and which is not.
- Remove the stop words and punctuations
- Convert the text data into vectors
Building a sms spam classification model
- Split the data into train and test sets
- Use Sklearn built-in classifiers to build the models
- Train the data on the model
- Make predictions on new data
Import the required packages
%matplotlib inline
import matplotlib.pyplot as plt
import csv
import sklearn
import pickle
from wordcloud import WordCloud
import pandas as pd
import numpy as np
import nltk
from nltk.corpus import stopwords
from sklearn.feature_extraction.text import CountVectorizer, TfidfTransformer
from sklearn.tree import DecisionTreeClassifier
from sklearn.model_selection import GridSearchCV,train_test_split,StratifiedKFold,cross_val_score,learning_curve
please note! You might find that I have reimported some of these packages again later in the article, it is just for ease of use if I ever have to use those code blocks again in future projects, you may omit those.😉
Loading the Dataset
data = pd.read_csv('dataset/spam.csv', encoding='latin-1')
data.head()
![df-head](/_next/image?url=%2F_next%2Fstatic%2Fmedia%2Fdfhead.23598c4d.png&w=1200&q=75)
Removing unwanted columns
From the above figure, we can see that there are some unnamed columns and the label and text column name is not intuitive so let's fix those in this step.
data = data.drop(["Unnamed: 2", "Unnamed: 3", "Unnamed: 4"], axis=1)
data = data.rename(columns={"v2" : "text", "v1":"label"})
data[1990:2000]
![pretty-df](/_next/image?url=%2F_next%2Fstatic%2Fmedia%2Fprettydf.02bf2574.png&w=1080&q=75)
now that the data is looking pretty, let's move on.
data['label'].value_counts()
# OUTPUT
ham 4825
spam 747
Name: label, dtype: int64
Preprocessing and Exploring the Dataset
If you are completely new to NLTK and Natural Language Processing(NLP) I would recommend checking out this short article before continuing. Introduction to Word Frequencies in NLP
# Import nltk packages and Punkt Tokenizer Models
import nltk
nltk.download("punkt")
import warnings
warnings.filterwarnings('ignore')
Build word cloud to see which message is spam and which is not
ham words are the opposite of spam in this dataset, 🤷♂️ yeah I also don't have any clue why it is so.
ham_words = ''
spam_words = ''
# Creating a corpus of spam messages
for val in data[data['label'] == 'spam'].text:
text = val.lower()
tokens = nltk.word_tokenize(text)
for words in tokens:
spam_words = spam_words + words + ' '
# Creating a corpus of ham messages
for val in data[data['label'] == 'ham'].text:
text = text.lower()
tokens = nltk.word_tokenize(text)
for words in tokens:
ham_words = ham_words + words + ' '
let's use the above functions to create Spam word cloud and ham word cloud.
spam_wordcloud = WordCloud(width=500, height=300).generate(spam_words)
ham_wordcloud = WordCloud(width=500, height=300).generate(ham_words)
#Spam Word cloud
plt.figure( figsize=(10,8), facecolor='w')
plt.imshow(spam_wordcloud)
plt.axis("off")
plt.tight_layout(pad=0)
plt.show()
![spam-word-cloud](/_next/image?url=%2F_next%2Fstatic%2Fmedia%2Fspamwc.358d4271.png&w=1920&q=75)
#Creating Ham wordcloud
plt.figure( figsize=(10,8), facecolor='g')
plt.imshow(ham_wordcloud)
plt.axis("off")
plt.tight_layout(pad=0)
plt.show()
![ham-word-cloud](/_next/image?url=%2F_next%2Fstatic%2Fmedia%2Fhamwc.ba050f62.png&w=1920&q=75)
from the spam word cloud, we can see that "free" is most often used in spam.
Now, we can convert the spam
and ham
into 0 and 1 respectively so that the machine can understand.
data = data.replace(['ham','spam'],[0, 1])
data.head(10)
![label-head](/_next/image?url=%2F_next%2Fstatic%2Fmedia%2Flabelhead.ffb864a6.png&w=1080&q=75)
Removing punctuation and stopwords from the messages
Punctuation and stop words do not contribute anything to our model, so we have to remove them. Using NLTK library we can easily do it.
import nltk
nltk.download('stopwords')
#remove the punctuations and stopwords
import string
def text_process(text):
text = text.translate(str.maketrans('', '', string.punctuation))
text = [word for word in text.split() if word.lower() not in stopwords.words('english')]
return " ".join(text)
data['text'] = data['text'].apply(text_process)
data.head()
![removed-stopwords](/_next/image?url=%2F_next%2Fstatic%2Fmedia%2Frmstopwords.89ff41a0.png&w=1080&q=75)
Now, create a data frame from the processed data before moving to the next step.
text = pd.DataFrame(data['text'])
label = pd.DataFrame(data['label'])
Converting words to vectors
we can convert words to vectors using either Count Vectorizer or by using TF-IDF Vectorizer.
TF-IDF is better than Count Vectorizers because it not only focuses on the frequency of words present in the corpus but also provides the importance of the words. We can then remove the words that are less important for analysis, hence making the model building less complex by reducing the input dimensions.
I have included both methods for your reference.
Converting words to vectors using Count Vectorizer
## Counting how many times a word appears in the dataset
from collections import Counter
total_counts = Counter()
for i in range(len(text)):
for word in text.values[i][0].split(" "):
total_counts[word] += 1
print("Total words in data set: ", len(total_counts))
# OUTPUT
Total words in data set: 11305
# Sorting in decreasing order (Word with highest frequency appears first)
vocab = sorted(total_counts, key=total_counts.get, reverse=True)
print(vocab[:60])
# OUTPUT
['u', '2', 'call', 'U', 'get', 'Im', 'ur', '4', 'ltgt', 'know', 'go', 'like', 'dont', 'come', 'got', 'time', 'day', 'want', 'Ill', 'lor', 'Call', 'home', 'send', 'going', 'one', 'need', 'Ok', 'good', 'love', 'back', 'n', 'still', 'text', 'im', 'later', 'see', 'da', 'ok', 'think', 'Ì', 'free', 'FREE', 'r', 'today', 'Sorry', 'week', 'phone', 'mobile', 'cant', 'tell', 'take', 'much', 'night', 'way', 'Hey', 'reply', 'work', 'make', 'give', 'new']
# Mapping from words to index
vocab_size = len(vocab)
word2idx = {}
#print vocab_size
for i, word in enumerate(vocab):
word2idx[word] = I
# Text to Vector
def text_to_vector(text):
word_vector = np.zeros(vocab_size)
for word in text.split(" "):
if word2idx.get(word) is None:
continue
else:
word_vector[word2idx.get(word)] += 1
return np.array(word_vector)
# Convert all titles to vectors
word_vectors = np.zeros((len(text), len(vocab)), dtype=np.int_)
for i, (_, text_) in enumerate(text.iterrows()):
word_vectors[i] = text_to_vector(text_[0])
word_vectors.shape
# OUTPUT
(5572, 11305)
Converting words to vectors using TF-IDF Vectorizer
#convert the text data into vectors
from sklearn.feature_extraction.text import TfidfVectorizer
vectorizer = TfidfVectorizer()
vectors = vectorizer.fit_transform(data['text'])
vectors.shape
# OUTPUT
(5572, 9376)
#features = word_vectors
features = vectors
Splitting into training and test set
#split the dataset into train and test set
X_train, X_test, y_train, y_test = train_test_split(features, data['label'], test_size=0.15, random_state=111)
Classifying using sklearn's pre-built classifiers
In this step we will use some of the most popular classifiers out there and compare their results.
Classifiers used:
- spam classifier using logistic regression
- email spam classification using Support Vector Machine(SVM)
- spam classifier using naive bayes
- spam classifier using decision tree
- spam classifier using K-Nearest Neighbor(KNN)
- spam classifier using Random Forest Classifier
We will make use of sklearn library. This amazing library has all of the above algorithms we just have to import them and it is as easy as that. No need to worry about all the maths and statistics behind it.
#import sklearn packages for building classifiers
from sklearn.linear_model import LogisticRegression
from sklearn.svm import SVC
from sklearn.naive_bayes import MultinomialNB
from sklearn.tree import DecisionTreeClassifier
from sklearn.neighbors import KNeighborsClassifier
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import accuracy_score
#initialize multiple classification models
svc = SVC(kernel='sigmoid', gamma=1.0)
knc = KNeighborsClassifier(n_neighbors=49)
mnb = MultinomialNB(alpha=0.2)
dtc = DecisionTreeClassifier(min_samples_split=7, random_state=111)
lrc = LogisticRegression(solver='liblinear', penalty='l1')
rfc = RandomForestClassifier(n_estimators=31, random_state=111)
#create a dictionary of variables and models
clfs = {'SVC' : svc,'KN' : knc, 'NB': mnb, 'DT': dtc, 'LR': lrc, 'RF': rfc}
#fit the data onto the models
def train(clf, features, targets):
clf.fit(features, targets)
def predict(clf, features):
return (clf.predict(features))
pred_scores_word_vectors = []
for k,v in clfs.items():
train(v, X_train, y_train)
pred = predict(v, X_test)
pred_scores_word_vectors.append((k, [accuracy_score(y_test , pred)]))
Predictions using TFIDF Vectorizer algorithm
pred_scores_word_vectors
# OUTPUT
[('SVC', [0.9784688995215312]),
('KN', [0.9330143540669856]),
('NB', [0.9880382775119617]),
('DT', [0.9605263157894737]),
('LR', [0.9533492822966507]),
('RF', [0.9796650717703349])]
Model predictions
#write functions to detect if the message is spam or not
def find(x):
if x == 1:
print ("Message is SPAM")
else:
print ("Message is NOT Spam")
newtext = ["Free entry"]
integers = vectorizer.transform(newtext)
x = mnb.predict(integers)
find(x)
# OUTPUT
Message is SPAM
Checking Classification Results with Confusion Matrix
If you are confused about the confusion matrix, read this small article before proceeding - The ultimate guide to confusion matrix in machine learning
from sklearn.metrics import confusion_matrix
import seaborn as sns
# Naive Bayes
y_pred_nb = mnb.predict(X_test)
y_true_nb = y_test
cm = confusion_matrix(y_true_nb, y_pred_nb)
f, ax = plt.subplots(figsize =(5,5))
sns.heatmap(cm,annot = True,linewidths=0.5,linecolor="red",fmt = ".0f",ax=ax)
plt.xlabel("y_pred_nb")
plt.ylabel("y_true_nb")
plt.show()
![confusion matrix](/_next/image?url=%2F_next%2Fstatic%2Fmedia%2Fcm.6aac9013.png&w=750&q=75)
from the confusion matrix, we can see that the Naive Bayes model is balanced. That's it !! we have successfully created a spam classifier. 🎉🎉
Thanks for the read!
You can find the code and dataset here.
If you liked this tutorial please do share it with your friends or on social media!
🎃 In case of any suggestions, corrections or if you want to have a chat about data science? Ping me on Twitter