Case file CS-01DSC790Master's dissertation
Machine Learning-Based Sentiment Analysis for an Intelligent Decision Support System in Malaysian Kopi F&B SMEs
A Master of Data Science dissertation that classifies sentiment in 1,000+ code-switched Malay-English customer reviews (Kopi Kenangan case study) and turns the results into a Streamlit decision-support app with an LLM business-analyst agent.
- Headline
- ≈92.5%accuracy, Logistic Regression + SMOTE
- Course
- DSC790 Data Science Project
- Format
- Master's dissertation
- Programme
- Master of Data Science, UiTM
- Supervisor
- Ts Dr Ahmad Faiz Mohd Ghazali
// 01 · Problem
The problem
Independent kopi F&B SMEs account for roughly 92% of Malaysia's 136,453 F&B establishments. Their customers leave reviews constantly, but the reviews are written in informal, code-switched Malay-English that is hard to read at scale.
The dissertation asks whether machine-learning sentiment analysis can turn those reviews into an intelligent decision support system (IDSS) for kopi F&B SMEs, using Kopi Kenangan as the case study.
// 02 · Approach
The approach
STEP 01
Review data
Analysed 1,000+ customer reviews from the Kopi Kenangan case study, written in informal, code-switched Malay-English.
STEP 02
Six classifiers
Compared Naive Bayes, SVM, Logistic Regression, Decision Tree, Random Forest and XGBoost on the same sentiment-classification task.
STEP 03
Five experimental conditions
Ran the models under baseline, hyperparameter-tuned, SMOTE-resampled and two ensemble-combination conditions, producing 20 model-condition results.
STEP 04
Decision-support app
Built a Streamlit decision-support app in which the open-source Qwen 2.5 7B LLM (Hugging Face) acts as an AI business-analyst agent.
// 03 · Results
The results
≈92.5%
Best accuracy
Logistic Regression with SMOTE
20
Model-condition results
Across 5 experiments
6
Classifiers compared
1,000+
Customer reviews analysed
- Logistic Regression with SMOTE was selected as the best model, reaching about 92.5% accuracy in sentiment classification.
- The work is delivered as a Streamlit decision-support app with Qwen 2.5 7B acting as an AI business-analyst agent.
- An extended abstract was prepared for PEERS 2026.
// 04 · Tools
Tools and methods
- Python
- Streamlit
- Hugging Face
- Qwen 2.5 7B
- SMOTE
- Logistic Regression
- Naive Bayes
- SVM
- Decision Tree
- Random Forest
- XGBoost
// 05 · Role & team
Role and team
Individual dissertation for the Master of Data Science at Universiti Teknologi MARA (UiTM).
SupervisorTs Dr Ahmad Faiz Mohd Ghazali