Machine Learning

Machine Learning (ML) is a transformative field within artificial intelligence that enables systems to learn from data, identify patterns, and make decisions or predictions without being explicitly programmed. As industries become increasingly data-driven, ML plays a pivotal role in powering intelligent applications across domains such as healthcare, finance, manufacturing, and environmental monitoring.

At the Academy of Technology (AOT), research in machine learning spans both foundational algorithm development and applied problem-solving. Faculty and students work on supervised, unsupervised, and reinforcement learning techniques, focusing on improving model accuracy, interpretability, and scalability. Research also explores deep learning architectures such as convolutional neural networks (CNNs), recurrent neural networks (RNNs), and transformer models for complex tasks like image recognition, natural language understanding, and time-series forecasting. AOT emphasizes practical applications that address real-world challenges. In healthcare, ML models are used for disease prediction, medical image analysis, and personalized treatment planning. In finance, predictive analytics and anomaly detection help in fraud prevention and risk assessment. In manufacturing, ML-powered predictive maintenance reduces downtime and improves operational efficiency. Environmental monitoring applications include climate pattern analysis and early warning systems for natural disasters.

Ethics, fairness, and transparency in machine learning are critical research considerations. AOT’s work includes explainable AI (XAI) techniques to ensure that decision-making processes are interpretable and trustworthy. Additionally, efforts are made to optimize models for deployment on resource-constrained devices, enabling intelligent capabilities in embedded systems and IoT applications. AOT’s machine learning research is supported by advanced computing infrastructure, GPU-enabled systems, and collaborative projects with industry partners. Students engage in hands-on experimentation using popular ML frameworks such as TensorFlow and PyTorch, gaining skills relevant to both academia and industry. Looking ahead, AOT aims to advance research in federated learning, self-supervised learning, and AI-driven automation, contributing to intelligent systems that are adaptive, ethical, and impactful.

Researchers:

Dr. Bappaditya Mondal (CSE Department)
Mr. Subhashis Das (CSE Department)
Mrs. Sarjita Soo (CSE Department)

Projects:

Phishing and Spam Detection Using Machine Learning Models
Sales Forecasting Using Sentiment Analysis of Customer
Detection of Phishing website and Browser Extension using Machine Learning techniques

Problem Description: Email and websites are two important mediums of communication in recent times. Technically, email and website-based communications rely on some specific uniform resource locator addresses. However, this type of communication suffers from the risk of cyber-attack. Hence, detection of phishing websites and classification of spam and ham emails are necessary for secure communication. One of the prime drawbacks in traditional approaches relates to detection and removal of constant and null values associated with uniform resource locators. Intruders often target and manipulate these constant and null values to make legitimate URLs into phished ones. Our aim is to propose an efficient approach compared to the traditional approaches to detect phishing websites and to classify spam and ham emails using machine learning techniques.

Focus of Research:

Focuses on enhancing the security of communication through emails and websites.
Aims to detect phishing websites and classify emails into spam and ham using machine learning techniques.

Publications:

Subrata Datta, Shaon Bandyopadhyay, Bappaditya Mondal, Classification of Spam and Ham Emails with Machine Learning Techniques for Cyber Security, IEEE International Conference on Integrated Intelligence and Communication Systems (ICIICS – 2023), Sharnbasva University, Karnataka, India. https:// 10.1109/ICIICS59993.2023.10421467
Bappaditya Mondal, Nayan Ranjan Das, Subrata Datta, Abhisek Saha, Trinetra Banerjee, URL phishing detection with exploratory data analysis for secure customer communication, 1st International Conference on Sustainable Communication, Machine Intelligence and Metaverse (SCMIM) – 2025, MCKVIE, Liluah, Howrah, India.
Tanmoy Chanda, Biva Mondal, Subhashis Das, Soumendu Banerjee, Bappaditya Mondal, A Deep Learning Based Approach for the Classification of Spam and Ham Emails, 2nd International Conference on Data Mining and Information Security (ICDMIS 2025), Eminent College of Management & Technology (ECMT), Kolkata, India. (Communicated)

Student Projects:

Title: Fake News Detection Using BiLSTM Deep Learning Model

Students: Sayantan Dutta, Rowmik Khanra, Subhadeep Paul, Abhishek Banerjee.

Supervisor: Dr. Bappaditya Mondal (CSE Department)

Year: 2024

Problem Description: Accurate sales forecasting is vital for planning and decision-making but remains challenging due to dynamic factors like market trends, competition, and customer behaviour. In online retail, customer feedback like ratings, reviews, and helpfulness votes offers key insights. Incorporating these into forecasting models can significantly enhance prediction accuracy and business performance.

Focus of Research:

Enhances the TSMixer model by integrating customer satisfaction indicators (ratings, reviews, helpfulness votes) into the sales forecasting pipeline.
Aims to capture the impact of customer sentiment on sales to improve forecasting accuracy.
Demonstrates significant error reduction (65%–99%) over existing models using Amazon datasets.

Publications:

P. Ghosh, S. Das, S. Roy, A. Bhattacharjee, A. Cortesi and S. Sen, SentiTSMixer: A Specific Model for Sales Forecasting Using Sentiment Analysis of Customer, IEEE Access, vol. 13, pp. 85882-85897, 2025. https:// 10.1109/ACCESS.2025.3570080. This work was supported in part by the Interconnected NordEst Innovation Ecosystem (iNEST) funded by PNRR (Mission 4.2, Investment 49 1.5) NextGeneration EU under Grant ECS_00000043—CUP H43C22000540006.

Problem Description: Phishing attacks are a growing digital threat, using fake websites or malicious browser extensions to steal sensitive data like passwords and banking details. Traditional detection methods, such as blacklists, often fail against rapidly evolving tactics and deceptive content. This project proposes a machine learning-based system to detect phishing threats more effectively. By analysing features like URL structure, domain age, page content, script behaviour, and extension permissions, the ML models aim to identify both phishing websites and harmful extensions. The objective is to enhance real-time protection and strengthen web safety for users worldwide.

Focus of Research:

How can machine learning models effectively differentiate between legitimate and phishing websites using structural, behavioural, and content-based features?
What features and indicators can be extracted from browser extensions to accurately detect potentially malicious behaviour in real-time?
How can a unified ML-based system be designed to ensure efficient and scalable detection of phishing threats across both websites and browser extensions?