Volume 9, Issue 12, December – 2024 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14609293

Spam Detection Using Large

Datasets with Multilingual Support
Anil Kumar Jatra1 (Student)
Master of Technology in Artificial Intelligence and Machine Learning

Kusum Sharma2 (Guide)

ORCID ID: 0009-0005-3220-159X
Department of Computer Science and Technology
RSR Rungta College of Engineering & Technology Kohka Kurud Road Bhilai
Chhattisgarh, India

Abstract:- Spam detection in the era of big data requires models, we aim to address the challenges posed by different
scalable and efficient techniques, particularly when languages. The goal is to build a system that is scalable and
dealing with large datasets containing diverse languages. accurate in detecting spam across various languages. The
Traditional methods struggle to address the multilingual paper covers designing, implementing, and testing such a
nature of spam, as language-specific approaches may not system, focusing on tasks like extracting features, processing
generalize well across different languages. This paper text, and adapting models to ensure effective multilingual
explores the establishment of a spam block method that spam detection.
leverages large, diverse datasets encompassing multiple
languages. We employ advanced machine-learning II. LITERATURE REVIEW
techniques to handle the complexities of linguistic
variations. By incorporating cross-lingual embeddings, This literature review summarizes various research
transfer learning, and ensemble models, our system aims works that investigates the utilization of machine learning
to detect spam content across various languages techniques for finding SMS spam, highlighting their methods,
accurately. We highlight the importance of feature datasets, results, and future scopes.
extraction, text preprocessing, and model adaptation in
achieving robust multilingual spam detection. The A. Studies on Machine Learning Models for SMS Spam
proposed approach demonstrates improved performance Detection
in detecting spam messages while maintaining scalability
and adaptability to new languages, providing a  SMS Spam Detection Using Naive Bayes and SVM
foundational framework for combating spam globally.
 Dataset: SMS Spam Collection Dataset, Kaggle datasets.
Keywords:- Spam Detection, Multilingual Spam, Machine  Findings: Naive Bayes and SVM demonstrate high
Learning, Cross-Lingual Embeddings, Transfer Learning, efficiency in handling high-dimensional data, achieving
Ensemble Methods, Feature Extraction, Text Preprocessing, accurate spam classification.
Model Adaptation, Large Datasets, and Language-  Future Scope: Real-time deployment using lightweight
Independent Spam Detection. frameworks like Flask.

I. INTRODUCTION  Performance of Random Forest and SVM

Spam detection is a crucial task, especially with the vast  Reference: Journal of Physics: Conference Series.
amount of data generated every day, including messages from  Findings: Random Forest and SVM performed robustly,
emails, social media, and messaging apps. Traditional achieving up to 95% accuracy. Preprocessing techniques
methods struggle to detect spam effectively when dealing like TF-IDF significantly improved classification results.
with multiple languages. Most existing spam detection  Future Scope: Expanding datasets and developing large-
systems are built for specific languages, which means they scale, standardized benchmarks.
don’t work well for others, reducing their accuracy in  Future Scope: Expansion to multilingual datasets and
identifying spam in diverse datasets. With communication exploration of deep learning methods.
platforms reaching a global audience, there’s a growing need
for systems that can detect spam in different languages,
 Relevance Vector Machine (RVM)
ensuring they are scalable and adaptable.
 Reference: Journal of Computational Analysis and
This research focuses on creating a spam detection
Applications.
system that can handle large datasets with multiple languages
using machine learning techniques. By using methods like
cross-lingual embeddings, transfer learning, and combining

IJISRT24DEC1820 www.ijisrt.com 2471

Volume 9, Issue 12, December – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14609293
 Findings: RVM outperformed other models, achieving an  Future Scope: Scaling blockchain integration for real-
F1 score of 97.6%. However, it required longer training time systems and improving algorithm efficiency.
times.
 Future Scope: Dataset expansion and advanced ensemble  Bio-Inspired Algorithms
methods for real-time applications.
 Reference: IEEE Access 2017.
 Hybrid Approaches Using Ensemble Techniques  Findings: Techniques like Artificial Bee Colony and
Cuckoo Search hold promise but remain underexplored
 Reference: IJNRD, Volume 9. for spam classification.
 Findings: KNN with Manhattan distance and Random  Future Scope: Optimization of these algorithms and
Forest achieved a 97.78% accuracy. hybrid implementations.
 Future Scope: Integration of neural networks and
exploration of fuzzy logic.  Deep Learning for Multilingual Spam Detection

 Optimizing SMS Spam Detection with Ensemble Learning  Reference: Hindawi Applied Computational Intelligence
and Soft Computing.
 Reference: Journal of Computer Networks.  Findings: SVM outperformed CNN with 99.6% accuracy
 Findings: SVM achieved the highest accuracy (98.57%) in SMS spam detection.
among classifiers. Ensemble methods enhanced  Future Scope: Incorporating ensemble methods and
prediction reliability. expanding experiments to larger datasets.
 Future Scope: Addressing class imbalance with advanced
techniques like SMOTE and expanding datasets for D. Evaluation Metrics and Dataset Limitations
multilingual support.
 Common Datasets Used
B. Advanced Techniques and Neural Network Applications
 UCI SMS Spam Collection, Kaggle, and other publicly
 Transformer-Based Embeddings available datasets dominated research efforts.
 Issues: Class imbalance (more ham messages than spam)
 Reference: Sensors 2023. and limited linguistic diversity.
 Findings: Combining GPT-3 embeddings with an
ensemble of classifiers obtained 99.91% accuracy.  Model Assessment using Metrics
 Future Scope: Applying the model to diverse datasets,
including non-English languages.  Accuracy, Precision, Recall, and F1 Score were the most
commonly used metrics.
 Hybrid CNN-LSTM Model  Challenge: Lack of standardized evaluation methods
across studies.
 Reference: Future Internet 2020.
 Findings: Achieved an accuracy of 98.37% in spam E. Summary and Recommendations
detection for English and Arabic SMS datasets. While traditional algorithms like machine learning such
 Future Scope: Enhancing framework functionalities for as the Support vector machine, Random Forest method, and
smishing and phishing detection. Naïve Bayes technique remain highly effective, advanced
techniques like hybrid CNN-LSTM models and transformer-
 Content-Based Neural Networks based embeddings have set new benchmarks in spam
detection. Future work can be done by:
 Reference: IJE Transactions B: Applications.
 Findings: Averaged Neural Network achieved 98.8%  Enlarging datasets to include diverse languages and
accuracy with robust preprocessing methods, including formats.
feature engineering for URLs and emojis.  Exploring advanced learning models like bio-inspired
 Future Scope: Expanding datasets and developing large- methods and deep-learning techniques.
scale, standardized benchmarks.  Enhancing real-time deployment efficiency through
lightweight and scalable frameworks.
C. Emerging Technologies and Future Directions
This Related work addresses the challenges and growth
 Blockchain Integration for Spam Detection in SMS spam detection, offering a foundation for further
exploration.
 Reference: IJISAE, 2024.
 Findings: Combining blockchain with machine learning
ensures data transparency and integrity while maintaining
high classification accuracy.

IJISRT24DEC1820 www.ijisrt.com 2472

Volume 9, Issue 12, December – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14609293
III. PROBLEM IDENTIFICATION  Feature Extraction:

 Challenges in Model Adaptability:  Extract features from the text, like word n-grams
(combining words), character-level n-grams, and
 Spam detection models require continuous updating to frequency-based features.
adapt to evolving spam techniques.
 Generalization across different languages, regions, and Methods like TF-IDF and word embeddings techniques
datasets is limited. is utilized in the presentation of texts that explains the
meaning and works across different languages.
 Handling Imbalanced Datasets:
 Model Development:
 Imbalanced datasets (more "ham" than "spam") pose
challenges for model performance.  Base Models: Build models using simpler algorithms,
such as Logistic Regression, for efficiency and Gradient-
 Preprocessing Variability: Boosting methods or the Random Forest technique to
seize complicated patterns.
 Effective preprocessing (e.g., tokenization, stop-word  Ensemble Model: Combine predictions from multiple
removal, feature extraction) is essential but varies across root models approaching methods such as voting and
studies, impacting model performance. weighted average to improve efficiency by combining
different models.
 Feature Engineering Complexity:
 Cross-Lingual Embeddings and Transfer Learning:
 Selection and optimization of feature parameters, such as
message length and word frequency, significantly  Use pre-trained multilingual word embeddings like
influence results. FastText or mBERT (multilingual BERT) to
understand relationships between words in different
 Real-Time Deployment: languages.
 Apply transfer learning by using models trained on one
 Many models lack real-time applicability due to language to help understand other languages.
computational or latency issues.
 Ensemble Model Implementation:
 Limited Use of Advanced Techniques:
 Combine outputs from different base models using
 Limited exploration of advanced methods such as deep techniques like Voting, Stacking, or Weighted
learning, hybrid approaches, and bio-inspired algorithms. Averaging to improve accuracy.
 Use ensemble models like Random Forest with stacking
 Evaluation Metrics and Consistency: or XG Boost with different base classifiers to enhance
performance.
 Lack of standardized evaluation metrics across studies,
leading to challenges in comparing results.  Evaluation of Models:

IV. PROPOSED METHOD  Performance metrics are used such as accuracy, precision,
F1-score, and Recall across different languages.
To detect spam in multiple languages, we proposed the  Cross-validation methods are implemented to check
techniques that utilizes machine learning, especially whether the model works well with different language
ensemble classifiers, to improve accuracy and scalability. data or not.

 Data Preprocessing & Gathering of Data:  Adaptation and Scalability:

 Gathering of data: Gather large datasets that contain  Continuously improve the system by adding more data
spam messages from different sources like emails, social and training on new languages.
media, and messaging apps.  Ensure the system works efficiently with large datasets
 Method of Preprocessing: the data is cleaned by and different languages.
eleminating unnecessary information, and noise, and
making sure the text is consistent across languages. This This method combines machine learning techniques,
includes tokenization (breaking text into parts), stemming especially ensemble models, to detect spam more accurately
(reducing words to their roots), and normalization and efficiently in multiple languages.
(making the text uniform).

IJISRT24DEC1820 www.ijisrt.com 2473

Volume 9, Issue 12, December – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14609293
V. BACKGROUND WORK ON  Checking performance metrics like accuracy, precision,
AN ENSEMBLE MODEL recall, and F1-score.

To detect spam in multiple languages, we propose a  Performing Hyperparameter Tunning

method that uses machine learning techniques, especially
ensemble models, to improve accuracy and scalability. In  Adjusting some settings to our model to improve its
Ensemble models by combining strengths of many algorithms efficiency.
for improving comprehensive performance and robustness in  The cross-validation method is used to find the best
SMS spam detection. parameters while avoiding overfitting.

 Data Collection  Model Deployment

 Collect messages labeled as spam or ham.  Implement a trained model to categorize messages as
 Use universal datasets like: spam or ham in real-time.
 SMS Spam Collection Dataset: Popular dataset with  Allow users to see the classification results and provide
labeled SMS messages. feedback.
 Kaggle Datasets: Platforms like Kaggle offer a variety of
datasets for spam detection. Ensure data is accurate and To understand the model's efficiency we can also
properly labeled. include insight tools.

 Method of Data Preprocessing VI. MODEL DESIGN

 Tokenization: Breaking down the messages or texts into  Datasets

smaller pieces. In SMS spam detection, datasets play an important role
 Cleaning of texts: unnecessary texts or words, in training the machine learning models. These datasets
punctuation, and special characters are removed. contain labeled messages contained as spam or ham. The
 Handle Missing Data: Ensure there are no gaps in the machine learning models learn patterns and characteristics
data; fix them if found. like word frequencies and text structures from this data to
 Vectorization process: the process of converting text into differentiate spam from non-spam messages. As new
numbers using methods like TF-IDF (word embeddings). messages are analyzed, the model applies these patterns to
predict if they are spam. Regularly updating the data enhances
 Feature Selection method the model’s accuracy and capability to adapt to new types of
Selection of useful features to help in identifying spam spam.
messages, like:
 Data Extraction
 The length of Message. Data extraction involves collecting datasets of messages
 Existence of certain spam-related texts and patterns. labeled as spam or ham. After cleaning the data we prepare it
 Occurrence of specific phrases or words. for analysis using methods like TF-IDF to extract features.
Then we split the datasets into two parts (training and testing).
 Selection of Models The testing data evaluates how well the model performs, and
For text classification we select a machine learning modifications are done to enhance its efficiency. Once
Techniques like: trained, the model is set up and categorizes new messages.

 Gradient Boosting Machines (GBM)  Data Visualization or Exploratory Data Analysis (EDA)
EDA requires analyzing datasets visuals and statistics to
 Naïve Bayes Algorithm
understand its key properties. This step uses charts like
 Decision Tree classification
histograms or scatter plots to identify patterns, trends, or
 Random Forests method anomalies in the data. It helps decide the best approaches.
 Support Vector Machines
 Ensemble Classifiers  Feature Engineering
 Logistic Regression method Feature engineering involves creating or selecting the
most useful information from the raw data to improve model
 Training of Models performance. This includes:

 We divide the datasets into two parts i.e. training and  Choosing relevant features (e.g., message length or
testing. specific keywords).
 Train the selected model using the training data.  Handling missing data.
 Converting text or categories into numerical values.
 Evaluation of Models  Scaling features to ensure consistency. The goal is to
provide the model with the most meaningful inputs for
 Checking the model on testing data. better predictions.

IJISRT24DEC1820 www.ijisrt.com 2474

Volume 9, Issue 12, December – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14609293
 Model Building  Accuracy: Overall correctness of the model.
Building an ensemble classifier requies training many  Precision: It shows the metrics that calculate how the
models and combined their predictions to obtained best model correctly gives predictions in positive terms.
performance as compared to other models. Ensemble  Recall: It is the metrics that calculate model positive
techniques like bagging (i.e. Random Forest), boosting (i.e. instances from the datasets.
Gradient Boosting), or stacking, leverage the performance of  F1-Score: These metrics help ensure the model reliably
diverse model to increase in classification accuracy. These identifies spam while minimizing errors and also show the
classifiers work by aggregating the predictions from multiple balance between recall and precision.
base learners, which may include decision trees, logistic
regression, or other algorithms. The ensemble calculates the  Predictions
likelihood of a message being spam or non-spam by The trained model then analyzes new messages and
integrating predictions from these models, often using divides them into spam or ham. It uses features and patterns
majority voting or weighted averages. This method is widely learned during training to assign a label or probability to each
used for tasks like spam detection because it provides high message. This step is crucial for real-time spam detection,
accuracy and robustness by reducing over-fitting and ensuring incoming messages are classified quickly to protect
leveraging diverse perspectives on the data. users from spam.

 Model Evaluation
Evaluating the model involves measuring how well it
classifies SMS messages. Metrics like:

Fig 1 Ensemble Model Development

IJISRT24DEC1820 www.ijisrt.com 2475

Volume 9, Issue 12, December – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14609293
VII. CONCLUSION [7]. Shafi’l Muhammad Abdulhamid, (Member, IEEE),
Muhammad Shafie Abd Latiff, Haruna Chiroma,
This research tackles the challenge of detecting spam in (Member, IEEE), Oluwafemi Osho, Gaddafi Abdul-
multiple languages by creating a system that is accurate, Salaam, Adamu I. Abubakar, (Member, IEEE), and
scalable, and efficient. It uses advanced machine learning Tutut Herawan, “A Review on Mobile SMS Spam
methods like Ensemble classifiers, cross-language Filtering Techniques”IEEE Access Published:
embeddings, transfer learning, and combining multiple February 13, 2017
models to improve performance on large and diverse datasets. [8]. Pavas Navaney, Ajay Rana, Gaurav Dubey, “SMS
Spam Filtering using Supervised Machine Learning
 Key Points Include: Algorithms” Conference Paper DOI:
10.1109/CONFLUENCE. 2018.8442564
 Better Multilingual Support: The system works well with [9]. Pradeep K.B, “Sms spam detection using machine
different languages, making spam detection more learning and deep learning techniques”, Published:
effective worldwide. May 2022
 Advanced Methods: Using modern techniques like deep [10]. B Sai Deepthi, K Sudheer Kumar, CH B M Swaroop,
learning and combining models, the system achieves high K Satya Sudheer, “Sms spam filtering using machine
accuracy and reliability. learning” JETIR, May 2024, Volume 11, Issue 5 Sixth
 Scalability and Real-Time Use: The system can handle International Conference on Computing
large datasets and adapt to new languages and changing Methodologies and Communication (ICCMC 2022)
spam patterns quickly. [11]. Mr. Ravi H. Gedam, Dr. Sumit Kumar Banchhor, “An
 Future Possibilities: The research suggests expanding Enhanced SMS Spam Detection Framework Using
datasets, using new technologies like blockchain, and Blockchain and Machine Learning” IJISAE, 2024,
exploring nature-inspired algorithms to improve spam Volume 12(22s), Pages 728–739
detection further. [12]. Samadhan Nagre, “Mobile SMS Spam Detection
using Machine Learning Techniques” 2018 JETIR
This study provides a strong foundation for global spam December 2018, Volume 5, Issue 12
detection, ensuring it works accurately across languages and [13]. Manas Ranjan Bishi, N Sardhak Manikanta, G Hari
stays adaptable to new challenges. Surya Bharadwaj, P Siva Krishna Teja, Dr G Rama
Koteswara Rao, “Optimizing SMS Spam Detection:
REFERENCES Leveraging the Strength of a Voting Classifier
Ensemble” IJISAE, 2024, Volume 12(3), Pages 2458–
[1]. Shreya Menthe, Kanish Rawal, Mrudula Hirave, 2469
A.J.Patil, “SMS spam detection using machine [14]. Ahmed Alzahrani, “Explainable AI-based Framework
learning” DOI: 10.17148/IJARCCE.2024. 13307 for Efficient Detection of Spam from Text Using an
[2]. Suparna DasGupta, Soumyabrata Saha, Suman Kumar Enhanced Ensemble Technique”, Engineering,
Das, “SMS spam detection using machine learning” Technology & Applied Science Research Volume 14,
Journal of Physics: Conference Series DOI: No. 4, 2024, Pages 15596-15601
10.1088/1742-6596/1797/1/012017 [15]. Shushanta Pudasainia, Aman Shakyaa, ∗, Sanjeeb
[3]. Ravi H Gedam, Sumit Kumar Banchhor,” Sms spam Prasad Pandeya, Prakriti Paudelb, Sunil Ghimirec,
detection using machine learning” Journal of Prabhat Ale, “SMS Spam Detection using Relevance
Computational Analysis and Applications Volume 33, Vector Machine” 3rd International Conference on
No. 4, 2024 Evolutionary Computing and Mobile Sustainable
[4]. Arpita Laxman Gawade, Sneha Sagar Shinde, Networks (ICECMSN 2023)
Samruddhi Gajanan Sawant, Rutuja Santosh [16]. Abdallah Ghourabi, Manar Alohaly, “Enhancing
Chougule, Mrs Almas Amol Mahaldar “A Research Spam Message Classification and Detection Using
Paper of SMS Spam Detection” 2024 IJNRD, Volume Transformer-Based Embedding and Ensemble
9, Issue 3-03-2024, ISSN: 2456-4184 | IJNRD.ORG Learning” Sensors 2023, Volume 23, Article
[5]. Harshit Kumar Simbal, Aaryan Sharma, Smriti 3861DOI: 10.3390/s23083861
Kumari, Gautam Kumar, Harshvardhan Kumar,” [17]. Abdallah Ghourabi, Mahmood A. Mahmood, Qusay
Spam Sms Classifier Using Machine Learning M. Alzubi, “A Hybrid CNN-LSTM Model for SMS
Algorithms” IJFMR240219483, Volume 6, Issue 2, Spam Detection in Arabic and English Messages”
March-April 2024 Future Internet 2020, Volume 12, Article 156 DOI:
[6]. Gregorius Airlangga, “Optimizing SMS Spam 10.3390/fi12090156
Detection Using Machine Learning: A Comparative [18]. Mr. E.Sankar, Y Y S Shekhar Babu, M.Tridev, “Sms
Analysis of Ensemble and Traditional Classifiers” spam detection using machine learning” International
Journal of Computer Networks, Architecture, and Journal of Scientific Research in Engineering and
High-Performance Computing, Volume 6, Number 4, Management Volume 7, Issue 4, April 2023
October 2024 DOI: 10.47709/cnahpc. v6i4.482

IJISRT24DEC1820 www.ijisrt.com 2476

Volume 9, Issue 12, December – 2024 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.5281/zenodo.14609293
[19]. Umair Maqsood, Saif Ur Rehman, Tariq Ali, Khalid
Mahmood, Tahani Alsaedi, Mahwish Kundi, “An
Intelligent Framework Based on Deep Learning for
SMS and e-mail Spam Detection” Hindawi Applied
Computational Intelligence and Soft Computing,
Volume 2023 DOI: 10.1155/2023/6648970
[20]. Suvarna M, Sanjeev J R, Kiran K, Ganjendran, “Sms
spam detection using machine learning” DOI:
10.17148/IARJSET.2024.11440
[21]. Nisha Wilvicta, Pradeep N, Tharun R, Mohammed
Tousif, “Sms spam detection using machine learning”
International Journal of Advances in Engineering
Architecture Science and Technology DOI: 12.2023
13677758/IJAEAST. 2023.10.0001
[22]. Humaira Yasmin Aliza, Kazi Aahala Nagary, Eshtiak
Ahmed, Kazi Mumtahina Puspita, Khadiza Akter
Rimi, Ankit Khater, Fahad Faisal, “A Comparative
Analysis of SMS Spam Detection Employing
Machine Learning Methods” Proceedings of the
[23]. Andrew Kipkebut, Moses Thiga, Elizabeth Okumu,
“Machine Learning Sms Spam Detection Model”
Kabarak University International Conference on
Computing and Information Systems, October 14–15,
2019
[24]. Samadhan M. Nagare, Pratibha P. Dapke, Syed
Ahteshamuddin Quadri, Sagar B. Bandal, Manasi
Ram Baheti, “A Review on Various Approaches on
Spam Detection of Mobile Phone SMS” International
Journal for Research in Engineering Applications &
Management (IJREAM) ISSN: 2454-9150, Volume 9,
Issue 2, May 2023
[25]. Luo GuangJun, Shah Nazir, Habib Ullah Khan, Amin
Ul Haq, “Spam Detection Approach for Secure Mobile
Message Communication Using Machine Learning
Algorithms” Hindawi Security and Communication
Networks, Volume 2020 DOI: 10.1155/2020/8873639

IJISRT24DEC1820 www.ijisrt.com 2477