Page 776
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026
An Optimized Hybrid Machine Learning and Deep Learning
Framework for Phishing Detection
Prof. Usha K
1
, Gowri Kannakatti², Chithra R², Boodalu Priya², Bhumika R², Gangamma²
Assistant Professor, Dept. of CSE, Jain Institute of Technology, Davangere, Karnataka, India¹
UG Students, Dept. of CSE, Jain Institute of Technology, Davangere, Karnataka, India²
DOI:
https://doi.org/10.51583/IJLTEMAS.2026.150500064
Received: 01 April 2026; Accepted: 06 April 2026; Published: 01 June 2026
ABSTRACT
Phishing remains one of the most persistent cybersecurity threats, targeting users through fraudulent emails,
websites, and evolving digital platforms. Although machine learning (ML) and deep learning (DL) techniques
have improved detection rates, existing models still face limitations such as poor adaptability to new attack
patterns, reliance on manual feature extraction, and lack of multilingual support. This paper reviews recent
approaches in phishing detection and identifies key gaps in current systems. Based on this analysis, a hybrid
framework is proposed that combines automated feature extraction, optimization techniques, and multilingual
capability. The proposed approach aims to enhance detection accuracy, robustness, and scalability in real-world
environments.
INTRODUCTION
Phishing attacks have grown significantly in complexity, moving beyond simple email scams to more advanced
threats across web platforms and blockchain-based systems. These attacks exploit both human behavior and
system vulnerabilities to gain access to sensitive information such as login credentials and financial data.
Traditional detection methods, such as rule-based systems and blacklists, are often ineffective against new and
unknown phishing techniques. In contrast, ML and DL models have shown improved performance by
identifying patterns in URLs, email content, and metadata. However, several challenges still exist, including
dependence on manually designed features, limited adaptability to new attack variations, and lack of support for
multiple languages.
This study focuses on analyzing existing research and proposing a more flexible and efficient hybrid detection
model.
LITERATURE REVIEW
This section summarizes and evaluates five recent studies related to phishing detection.
Several studies highlight the transition from traditional rule-based systems to intelligent ML and DL-based
approaches. These modern techniques improve detection accuracy but often struggle with unseen phishing
attacks and require large datasets for training.
Deep learning-based models combined with optimization techniques have demonstrated high performance by
automatically extracting features and improving classification accuracy. However, such models may require
high computational resources and may not generalize well across different datasets.
Some research has explored multilingual phishing detection using machine learning and open-source
intelligence (OSINT). These approaches enhance detection in diverse linguistic environments but are limited by
dataset size and translation issues.
Page 777
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026
Other studies focus on phishing threats in emerging domains such as blockchain systems. While these works
provide valuable insights into new attack vectors, they do not offer complete automated detection solutions.
Feature selection techniques combined with deep learning models have also been used to reduce computational
complexity while maintaining reasonable accuracy. However, these approaches may not perform consistently
across different datasets.
Overall, existing research demonstrates progress in phishing detection but still leaves room for improvement in
terms of adaptability, efficiency, and real-time implementation.
Problem Statement
Despite advancements in phishing detection, several limitations remain:
1. Heavy reliance on manual feature engineering
2. Difficulty in detecting new and zero-day attacks
3. Limited support for multilingual and cross-domain scenarios
4. Imbalanced and outdated dataset
5. High computational cost for complex deep learning models The main
research question addressed in this paper is:
How can a phishing detection system be designed to be accurate, scalable, and adaptable while integrating
automated feature extraction and multilingual capabilities?
PROPOSED METHODOLOGY
To address the identified challenges, this paper proposes a Hybrid Phishing Detection Framework (HPDF).
Data Collection
Data is gathered from multiple sources such as PhishTank, OpenPhish, and publicly available datasets.
Multilingual datasets are also included to improve generalization.
Data Preprocessing
Preprocessing steps include cleaning the data, normalizing features, and handling class imbalance using
techniques such as SMOTE.
Feature Extraction
Instead of manual feature engineering, automated methods are used: Variational
Autoencoders (VAE) for deep feature extraction
OSINT-based features such as domain information, IP addresses, and network attributes
Model Design
The proposed hybrid model integrates multiple techniques: Convolutional Neural
Networks (CNN) for identifying URL patterns Long Short-Term Memory (LSTM)
networks for sequential data analysis Random Forest for ensemble-based classification
Page 778
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026
Optimization
Hyperparameter tuning is performed using optimization techniques such as grid search to improve model
performance.
Evaluation Metrics
The model is evaluated using standard metrics including:
Accuracy Precision Recall
F1-score
ROC-AUC
Experimental Setup
The system is implemented using Python with libraries such as TensorFlow, Keras, and Scikit-learn. A GPU-
enabled environment is used for efficient training. The dataset is divided into training, validation, and testing
sets in a 70:15:15 ratio. Cross-validation is applied to ensure reliability of results.
RESULTS AND DISCUSSION
The proposed hybrid model is expected to achieve high accuracy and improved detection performance compared
to individual ML or DL models. By combining multiple techniques, the system reduces false positives and
enhances the detection of previously unseen phishing attacks.
The use of automated feature extraction reduces dependency on human intervention, while optimization
techniques improve overall efficiency. Additionally, incorporating OSINT features enhances contextual
understanding of phishing behavior.
Future Work
Future improvements may include:
Page 779
www.rsisinternational.org
INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,
MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)
ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue V, May 2026
Real-time phishing detection systems Integration with browser
security tools
Use of explainable AI for better transparency
Expansion to emerging domains such as IoT and blockchain Application of federated
learning for privacy preservation
CONCLUSION
This paper reviewed existing phishing detection techniques and identified their limitations. A hybrid framework
was proposed to address these challenges by combining machine learning, deep learning, feature automation,
and optimization strategies. The proposed system aims to provide a scalable and effective solution for modern
phishing detection problems.
REFERENCES
1. S. Ahmad et al., “Across the Spectrum In-Depth Review AI-Based Models for Phishing Detection,”
IEEE Access, 2025.
2. K. Barik, S. Misra, and R. Mohan, “Web-based phishing URL detection model using deep learning
optimization techniques,” Int. J. Data Sci. Anal., 2025.
3. P. An et al., “Multilingual Email Phishing Attacks Detection using OSINT and Machine Learning,”
arXiv, 2025.
4. M. Qi et al., “EIP-7702 Phishing Attack,” arXiv, 2025.
5. G. S. Nayak et al., “Enhancing Phishing Detection: A Machine Learning Approach With Feature
Selection and Deep Learning Models,” IEEE Access, 2025.
