CogD-ML: Cognitive Determinant Analysis of Purchase Intention 🛒
Using Machine Learning and NLP 🤖

📚 [GitHub Repository] | 📝 [Paper - Under Review] | 🚀 [Demo Application]

Project Overview

CogD-ML is an integrated research framework that utilizes the Theory of Planned Behavior (TPB) alongside Machine Learning (ML) and Natural Language Processing (NLP) to predict e-commerce purchase intentions in developing economies. By analyzing survey data from 352 consumers in Bangladesh, the project identifies how cognitive factors—such as attitude and social pressure—drive digital shopping behavior.

Key Features

• Psychological Grounding: Built on the established TPB framework (Attitude, Subjective Norm, and Perceived Behavioral Control).
• Predictive Performance: Features a Random Forest model with 94% accuracy and a high F1-score of 0.965 .
• Explainable AI: Utilizes SHAP and LIME to rank cognitive determinants, revealing that peer encouragement (SN4) is a primary driver of intention.
• NLP Integration: Benchmarks Large Language Models (LLMs) like Llama 3.1 and Kimi-K2 for zero-shot and few-shot tabular classification.
• Robustness Testing: Includes threshold sensitivity analysis and temporal validation to ensure model stability across different time periods.

Methodology

The pipeline consists of the following technical stages:

1. Data Acquisition: Survey collection focusing on university students and young professionals.
2. Preprocessing: Label encoding, synthetic oversampling (SMOTE), and Cronbach's alpha reliability checks .
3. Statistical Analysis: Hypothesis testing via Structural Equation Modeling (SEM) and bivariate regression.
4. Model Suite: Training and evaluation of SVM, AdaBoost, kNN, Gradient Boosting, and Multilayer Perceptron (MLP).
5. Deployment: Implementation of a cloud-based predictor for real-time marketing insights.

Performance Results

The following table highlights the performance of the core classifiers evaluated in the study:

Model	Accuracy	F1-Score	Precision	Recall
Random Forest	94.0%	0.965	0.945	0.986
SVM	93.8%	0.963	0.947	0.980
AdaBoost	92.6%	0.956	0.953	0.959
Logistic Regression	92.6%	0.956	0.950	0.963

Citations

If you use it in your research or project, please cite it as follows:

@article{barua2025cogd,
title={CogD-ML: Cognitive Determinant Analysis of Purchase Intention Using Machine Learning and NLP},
author={Barua, Saikat and Rahman, Raiyan and Azad, Abul Kalam Al and Momen, Sifat},
journal={PLOS ONE},
year={2025},
note={Under review/submitted manuscript},
url={https://github.com/your-repo-link}
}