Citation

Aktar, Shuly and Majedul Huq, Sheikh and Kumar, Gyanendra and Gopal, R. and Afroz, Sonia and Ali, Md. Akkas (2025) AI-Powered Product Purchase Prediction: A Multi-Modal Deep Learning Approach. In: 5th International Conference on Emerging Research in Electronics, Computer Science and Technology, ICERECT 2025, 12 September 2025 - 13 September 2025, Mandya.

Text IEEE Xplore Full-Text PDF_12.pdf - Published Version Restricted to Repository staff only Download (1MB)

Abstract

AI and digital platforms have changed how practical analysis has become in sectors like cosmetics. Consumer perception towards cosmetic brands is crucial in making purchase decisions in Bangladesh, but traditional methods of measuring lack the reality. However, no studies have been found on modeling trust in such a domain in the context of cosmetic buying behavior in Bangladesh. Modern methods commonly focus either on structured data or free text, where there is a lack of the combination, and often, they do not provide interpretability that poses an obstacle to business utilization. The purpose of this study is to propose a method for predicting customer trust in cosmetics using an AI-based approach. It combines structured demographic and behavioral data with unstructured textual reviews; uses CatBoost and DNN for classification learning of models by SHAP and LIME to increase interpretability, and checks deployment readiness. The approach is a mixture of a structured response survey (Sample size: 1000) and unstructured review text analyzed through BERT embeddings. CatBoost and DNN models are adopted to predict trust toward consumers, as well as purchase intention. Model effectiveness is evaluated by measures such as accuracy, Precision, Recall, F1-score, MAE, MSE, RMSE, and R2. SHAP and LIME are used to increase the transparency and interpretability. We have implemented the DL models that outperform the baseline models with accuracy (93%, 90%), Precision (93%, 90%), Recall (92%, 87%) F1 -Score (93%, 89%), MAE (0.047, 0.036), MSE (0.004, 0.002), RMSE (0.059, 0.045), and R² score (98%, 92%) for CatBoost, and DNN, respectively for the purchase intention. SHAP identified essential factors related to trust, such as brand trust and review sentiments. The statistical analysis (ANOVA and Friedman test) supported the validity of the model. This AI-based methodology provides valuable results to cosmetic marketers operating in Bangladesh, presenting personalized strategies and better consumer engagement plans. Structured and unstructured data combined allow artificial intelligence in cosmetics to predict customer trust for scalable decision-making

Item Type:	Conference or Workshop Item (Paper)
Uncontrolled Keywords:	AI, Consumer Trust, CatBoost
Subjects:	Q Science > Q Science (General) > Q300-390 Cybernetics
Divisions:	Faculty of Information Science and Technology (FIST)
Depositing User:	Ms Rosnani Abd Wahab
Date Deposited:	20 Apr 2026 03:30
Last Modified:	20 Apr 2026 03:30
URII:	http://shdl.mmu.edu.my/id/eprint/15770

Downloads

Edit (login required)