Alamein University Data Mining

Faculty of Computer Science Course code: AIE323

& Engineering

PROJECTS IN THIS BOOKLET

Project 1 — Customer Segmentation using Clustering
Project 2 — Market Basket Analysis & Association Rule Mining
Project 3 — E-Commerce Behavioral Clustering & Customer Profiling
BONUS Project — Multi-Domain Sentiment Analysis & Opinion Mining

Project Instructions:

1- Each team must consist of 4 to 5 students (maximum).

2- Implement a complete pipeline, including:

• Data collection (or using a suitable dataset)

• Data cleaning & preprocessing
• Exploratory Data Analysis (EDA)
• Feature engineering/selection
• Modeling
• Evaluation

3- Deliverables:

1- GitHub Repository

• Upload all code in a well-organized notebook (Jupyter).

• Code should be clean, commented, and reproducible.

2- Application (Required)

• Build an interactive app using Dash or Streamlit to demonstrate your project.

3- Final Report (The report must include):

• Project overview and objectives

• Dataset description
• Detailed explanation of all steps
• Algorithms used and why they were chosen
• Results and evaluation
• Challenges/problems faced and how you solved them
 Alamein University Data Mining
Faculty of Computer Science Course code: AIE323
& Engineering

Project 1: Customer Segmentation using Clustering

Project Overview
The Customer Segmentation using Clustering project focuses on discovering natural groupings within customer
data using unsupervised learning techniques. will apply data mining methods — including K-Means, DBSCAN, and
Hierarchical Clustering — to segment customers based on behavioral, demographic, and transactional features.
The final delivery is an interactive Streamlit or Dash dashboard that lets stakeholders explore the discovered
segments visually and in real time.

Milestone 1: Data Collection, Exploration, and Preprocessing

Objectives
• Acquire a customer dataset and perform thorough exploration analysis to understand the structure and
quality.

Tasks
1. Data Collection:
◦ Acquiring a dataset from Kaggle, UCI, or a company CRM. Suggested datasets: Online Retail II, Mall
Customer Segmentation, or E-Commerce dataset.
◦ Ensure the dataset contains customer behavioral features such as purchase frequency, recency,
monetary value (RFM features), demographics, or product preferences.
2. Data Exploration:
◦ Conduct EDA to understand distributions, detect outliers, and examine feature relationships.
◦ Compute RFM (Recency, Frequency, Monetary) metrics if not already present.
3. Preprocessing:
◦ Handle missing values via imputation or removal. Remove duplicates.
◦ Normalize or standardize numerical features (essential for distance-based clustering).
◦ Encode any categorical variables using label or one-hot encoding.
4. EDA Visualizations:
◦ Generate histograms, box plots, correlation heatmaps, and pair plots to understand feature distributions.
◦ Visualize RFM distributions and identify preliminary customer groups.

Deliverables
EDA Report: A document summarizing key insights, data quality findings, and preprocessing decisions.
Interactive EDA Notebook: A Jupyter notebook with visualizations revealing key patterns.
Cleaned Dataset: A preprocessed, normalized dataset ready for clustering.

Milestone 2: Advanced Analysis and Feature Engineering

Objectives
• Enhance the dataset with engineered features and apply dimensionality reduction to prepare for clustering.

Tasks
5. Feature Engineering:
◦ Construct RFM scores and segment scores from raw transaction data.
 Alamein University Data Mining
Faculty of Computer Science Course code: AIE323
& Engineering
◦ Create behavioral features such as average basket size, product category diversity, and purchase time
patterns.
6. Dimensionality Reduction:
◦ Apply PCA (Principal Component Analysis) to reduce high-dimensional feature spaces while preserving
variance.
7. Statistical Analysis:
◦ Use correlation analysis and variance inflation factor (VIF) to remove redundant features.
◦ Compute silhouette scores for preliminary k values to guide model selection.
8. Data Visualization:
◦ Build heatmaps showing feature importance per preliminary cluster.

Deliverables
Feature Engineering Report: Documentation of all engineered features, transformation rationale, and expected
clustering impact.
Dimensionality Reduction Visualizations: 2D/3D scatter plots showing cluster structure before modeling.

Milestone 3: Clustering Model Development and Evaluation

Objectives
• Apply, compare, and evaluate multiple clustering algorithms to find the optimal customer segmentation.

Tasks
9. Algorithm Selection:
◦ Implement at least three clustering algorithms: K-Means, DBSCAN, and Agglomerative (Hierarchical)
Clustering.
◦ Optionally explore Gaussian Mixture Models (GMM) or BIRCH for large datasets.
10. Optimal Cluster Discovery:
◦ For K-Means: Use the Elbow Method and Silhouette Analysis to determine optimal k.
◦ For DBSCAN: Tune epsilon and min_samples using k-distance graphs.
◦ For Hierarchical Clustering: Use dendrograms to decide cut-off levels.
11. Cluster Evaluation:
◦ Compute Silhouette Score, Davies-Bouldin Index, and Calinski-Harabasz Score.
◦ Validate cluster stability using bootstrapping or cross-validation.
12. Cluster Profiling:
◦ Analyze each cluster's characteristics — describe the 'persona' for each group.
◦ Generate radar charts, parallel coordinate plots, and bar comparisons per cluster.

Deliverables
Clustering Report: Detailed comparison of all algorithms, including evaluation metrics, optimal parameters, and
final segment profiles.
Model Code: Python code for all clustering implementations, tuning, and evaluation.
Segment Profiles: A clear persona description for each customer discovered.

Milestone 4: Interactive Dashboard Deployment (Streamlit / Dash)

Objectives
• Build and deploy an interactive web application that allows stakeholders to explore customer segments.
 Alamein University Data Mining
Faculty of Computer Science Course code: AIE323
& Engineering
Tasks
13. Dashboard Development:
◦ Build the app using Streamlit or Dash. The dashboard must include: cluster visualization (2D/3D scatter),
cluster comparison bar charts, and a customer lookup tool (enter customer ID and get their segment).
◦ Include interactive filters for algorithm selection, number of clusters, and feature axes.
◦ Display RFM distributions, persona summaries, and actionable recommendations per segment.
14. Model Integration:
◦ Load the trained clustering model (saved with joblib or pickle) inside the app.
◦ Allow users to upload new customer data (CSV) and get real-time cluster assignments.
15. Deployment:
◦ Deploy locally and optionally to Streamlit Community Cloud, Heroku, or Render.
◦ Document all deployment steps, dependencies ([Link]), and configuration.

Deliverables
Deployed Dashboard: A live Streamlit or Dash app with full cluster exploration features.
Deployment Report: Setup instructions, architecture diagram, and screenshots of key dashboard views.

Milestone 5: Final Documentation and Presentation

Deliverables
Final Project Report: Full project summary: problem, data, methods, results, business insights, and future work.

Milestones Summary
Milestone Key Deliverables
1. Data Collection & EDA EDA Report, Interactive Visualizations, Cleaned Dataset

2. Feature Engineering & Reduction Feature Engineering Report, PCA Visualizations

3. Clustering Model Development Clustering Report, Model Code, Segment Profiles

4. Dashboard Deployment Deployed Streamlit/Dash App, Deployment Report

5. Final Documentation Final Project Report.

 Alamein University Data Mining
Faculty of Computer Science Course code: AIE323
& Engineering

Project 2: Market Basket Analysis & Association Rule Mining

Project Overview
The Market Basket Analysis project applies association rule mining to discover hidden purchasing patterns in
transactional retail data. Students will implement the Apriori and FP-Growth algorithms to uncover frequent
itemsets and generate meaningful rules (e.g., 'customers who buy bread also buy butter'). The insights will be
presented through an interactive Streamlit or Dash dashboard that allows business teams to explore rules, filter
by support and confidence thresholds, and generate product recommendations in real time.

Milestone 1: Data Collection, Exploration, and Preprocessing

Objectives
• Collect transactional retail data and prepare it in the correct basket format for association mining.

Tasks
16. Data Collection:
◦ Acquire transactional datasets from sources such as Kaggle (Online Retail II, Groceries Dataset,
Instacart) or UCI Repository.
◦ Ensure the dataset contains transaction IDs, product/item names, and quantities.
17. Data Exploration:
◦ Perform EDA to understand product frequency distributions, transaction sizes, and seasonal trends.
◦ Identify top-selling items, rare items, and potential noise in the catalog.
18. Data Preprocessing:
◦ Handle missing values, remove cancelled transactions, and clean item descriptions.
◦ Transform the dataset into a binary transaction matrix (basket format) required by association algorithms.
◦ Apply minimum frequency filtering to remove extremely rare items that add noise.
19. EDA Visualizations:
◦ Create item frequency bar charts, transaction size histograms, and word clouds of product names.
◦ Visualize product co-occurrence matrices as heatmaps.

Deliverables
EDA Report: Summary of data distribution, quality issues, and preprocessing steps.
Transaction Matrix: A clean binary basket matrix ready for association rule mining.
EDA Notebook: Jupyter notebook with all visualizations.

Milestone 2: Advanced Analysis and Pattern Discovery

Objectives
• Perform deep analysis on co-purchase patterns and identify preliminary associations before formal mining.

Tasks
20. Co-occurrence Analysis:
◦ Build and visualize product co-occurrence matrices to preview strong relationships.
◦ Identify cross-category purchasing patterns (e.g., produce with dairy).
21. Statistical Measures:
◦ Compute pairwise lift, chi-squared, and Jaccard similarity for top product pairs.
 Alamein University Data Mining
Faculty of Computer Science Course code: AIE323
& Engineering
◦ Analyze which categories drive the most cross-selling opportunities.
22. Segmentation by Basket:
◦ Group customers by basket composition using simple clustering before rule mining.
23. Visualization:
◦ Develop network graphs showing product relationships (nodes = items, edges = co-purchases).
◦ Build interactive dashboards to explore product category interactions.

Deliverables
Co-occurrence Analysis Report: Statistical analysis of item relationships and preliminary findings.
Network Graph: An interactive product relationship network visualization.

Milestone 3: Association Rule Mining — Apriori & FP-Growth

Objectives
• Implement Apriori and FP-Growth algorithms, mine frequent itemsets, and generate actionable rules.

Tasks
24. Algorithm Implementation:
◦ Implement Apriori using the mlxtend library. Experiment with different minimum support thresholds (e.g.,
0.01, 0.02, 0.05).
◦ Implement FP-Growth for efficient mining on larger datasets. Compare performance with Apriori.
25. Rule Generation:
◦ Generate association rules with metrics: Support, Confidence, and Lift.
◦ Filter rules by minimum lift > 1.2 and confidence > 0.5 for actionable results.
◦ Optionally compute Conviction and Leverage as additional rule strength measures.
26. Rule Analysis:
◦ Rank rules by lift to identify the strongest associations.
◦ Group rules by antecedent category (e.g., all rules where bread is an antecedent).
◦ Detect redundant rules and apply pruning strategies.
27. Evaluation:
◦ Compare Apriori vs FP-Growth on runtime performance and rule quality.
◦ Validate discovered rules with domain knowledge or manual inspection.

Deliverables
Mining Report: Full analysis of frequent itemsets and rules, algorithm comparison, and business interpretations.
Rule Database: A curated CSV/Excel file of the top rules with all metrics.
Algorithm Code: Python notebooks for both Apriori and FP-Growth implementations.

Milestone 4: Recommendation Engine & Dashboard Deployment

Objectives
• Build a rule-based product recommendation engine and deploy it as an interactive Streamlit or Dash
application.

Tasks
28. Recommendation Engine:
◦ Build a function given a set of items in a cart, return the top-N recommended items ranked by lift.
 Alamein University Data Mining
Faculty of Computer Science Course code: AIE323
& Engineering
◦ Handle multi-item antecedents and edge cases (no rules found, cold-start problem).
29. Streamlit / Dash App Features:
◦ MUST include: a Rule Explorer (searchable, filterable table of all mined rules).
◦ MUST include: a Product Recommender (user selects items in a basket and gets top recommendations).
◦ MUST include: a Network Graph tab showing product association clusters interactively (use Plotly or
Pyvis).
◦ MUST include threshold sliders for support, confidence, and lift to dynamically update displayed rules.
30. Deployment:
◦ Deploy locally and optionally on Streamlit Cloud, Render, or a similar platform.
◦ Provide a [Link] with setup instructions.

Deliverables
Deployed App: A live recommendation and rule exploration app built with Streamlit or Dash.

Milestone 5: Final Documentation and Presentation

Deliverables
Final Project Report: End-to-end documentation covering methodology, rules discovered, business impact, and
recommendations.

Milestones Summary
Milestone Key Deliverables
1. Data Collection & Preprocessing EDA Report, Transaction Matrix, EDA Notebook

2. Co-occurrence & Pattern Analysis Co-occurrence Report, Network Graph Visualization

3. Apriori & FP-Growth Mining Mining Report, Rule Database, Algorithm Code

4. Recommendation Engine & Dashboard Deployed Streamlit/Dash App

5. Final Documentation & Presentation Final Project Report, Final Presentation

 Alamein University Data Mining
Faculty of Computer Science Course code: AIE323
& Engineering

Project 3: Healthcare Patient Clustering & Medical Risk

Grouping
Project Overview
The Healthcare Patient Clustering & Medical Risk Grouping project applies unsupervised learning to real-
world clinical data to automatically discover meaningful patient subgroups. Rather than relying on
predefined diagnoses, students will let the data reveal hidden patterns across lab results, vital signs,
demographics, and medical histories. Discovered clusters will be clinically interpreted as risk profiles (e.g.,
'High-Risk Diabetic Group', 'Stable Elderly Group') and deployed as an interactive Streamlit or Dash clinical
decision-support dashboard for healthcare professionals.

Milestone 1: Data Collection, Exploration, and Medical Preprocessing

Objectives
• Acquire clinical datasets, perform domain-aware EDA, and build a medically sound preprocessing pipeline.
Tasks
1. Data Collection:
◦ Acquire a clinical dataset from open sources. Recommended: MIMIC-III (PhysioNet), Pima Indians
Diabetes (Kaggle), Heart Disease UCI, or the chronic kidney disease dataset.
◦ Ensure features cover at least three of: lab results (glucose, creatinine, HbA1c), vital signs (BP, BMI, heart
rate), demographics (age, sex), and diagnoses/medications.
2. Medical Data Exploration:
◦ Perform EDA with clinical awareness: identify which features are continuous (lab values), binary
(diagnosis flags), or ordinal (severity scores).
◦ Examine missing data patterns — in clinical data, missingness is often clinically
◦ Compute summary statistics stratified by age group, sex, and primary diagnosis.
3. Medical Preprocessing:
◦ Handle missing values using clinically appropriate strategies.
◦ Detect and handle clinical outliers carefully — an extreme lab value may be a real critical result, not
noise.
◦ Normalize continuous features (min-max or z-score) to prevent features with large ranges (e.g., creatinine
vs. age) from dominating distance calculations.
◦ Encode categorical variables (diagnosis codes, medication classes) using binary or ordinal encoding.
4. EDA Visualizations:
◦ Create clinical distribution plots: violin plots for lab values per age group, heatmaps of feature
correlations, and missing data pattern matrices.
◦ Plot comorbidity co-occurrence networks: which conditions appear together most frequently.

Deliverables:

Medical EDA Report: Summary of feature distributions, clinical patterns, missingness analysis, and
preprocessing decisions.
Comorbidity Network: Visualization of co-occurring conditions across the patient cohort.
Cleaned Clinical Dataset: A fully preprocessed, normalized dataset ready for clustering.

Milestone 2: Clinical Feature Engineering and Dimensionality Analysis

Objectives
• Engineer medically meaningful composite features and prepare the feature space for robust clustering.
Tasks
5. Clinical Feature Engineering:
◦ Computer composite risk scores: eGFR (estimated Glomerular Filtration Rate) from creatinine and age,
BMI from height and weight, cardiovascular risk proxies.
 Alamein University Data Mining
Faculty of Computer Science Course code: AIE323
& Engineering
◦ Create binary flags for clinically defined thresholds (e.g., HbA1c >= 6.5 = diabetic range, SBP >= 140 =
hypertensive).
◦ Build a multi-morbidity score: count of concurrent chronic conditions per patient.
◦ Engineer age-group bins (paediatric, adult, elderly) and interaction features (age × BMI, age × creatinine).
6. Feature Selection:
◦ Use variance thresholding to remove near-zero-variance features.
◦ Apply mutual information, correlation filtering (drop features with |r| > 0.9), and domain expert rules to
select a final feature set.
7. Dimensionality Reduction:
◦ Apply PCA and examine explained variance ratios. Select components that explain at least 80% of the
variance.
Deliverables:

Clinical Feature Engineering Report: Documentation of all composite scores, threshold flags, and feature
selection rationale.
PCA & Projection Visualizations: Explained variance plots.

Milestone 3: Clustering Model Development and Clinical Validation

Objectives
• Apply and compare multiple clustering algorithms, then validate and clinically interpret discovered patient
groups.
Tasks
8. Algorithm Implementation:
◦ Implement K-Means with Elbow Method and Silhouette Analysis for optimal k selection.
◦ Implement Agglomerative Hierarchical Clustering with Ward, Complete, and Average linkage. Use
dendrograms to guide cluster selection.
◦ Implement DBSCAN to detect high-density patient subgroups and identify clinical outliers (anomalous
patients).
◦ Implement Gaussian Mixture Models (GMM) to capture soft/probabilistic cluster memberships —
clinically useful as a 'risk probability' per patient.
9. Algorithm Evaluation:
◦ Compute Silhouette Score, Davies-Bouldin Index, and Calinski-Harabasz Score for each algorithm.
◦ Compare runtime and stability across algorithms using bootstrapped subsamples.
10. Clinical Cluster Interpretation:
◦ For each discovered cluster, compute mean/median values of all clinical features and compare against
population norms.
◦ Conduct ANOVA or Kruskal-Wallis tests to confirm clusters are statistically distinct on key clinical
variables.
◦ Name each cluster with a clinically meaningful label (e.g., 'Early-Stage Metabolic Risk', 'Severe Renal
Impairment', 'Healthy Low-Risk').
◦ Build risk stratification tiers: assign Low / Medium / High risk labels to clusters based on clinical feature
profiles.
11. Survival or Outcome Analysis (if data permits):
◦ If the dataset contains outcome labels (readmission, mortality, disease progression), compare outcome
rates across clusters using Kaplan-Meier curves or chi-squared tests.
Deliverables
Clustering Comparison Report: Full algorithmic comparison with evaluation metrics, dendrogram, and
justification for the chosen model.
Clinical Cluster Profiles: Statistical and clinical characterization of each patient group with risk tier
assignments.
Model Code: Python code for all clustering implementations, evaluation, and profile analysis.

Milestone 4: Clinical Decision-Support Dashboard (Streamlit / Dash)

Objectives
 Alamein University Data Mining
Faculty of Computer Science Course code: AIE323
& Engineering
• Deploy a clinical-grade interactive dashboard that allows healthcare professionals to explore patient clusters
and assess individual patient risk.
Tasks
12. Dashboard Pages (All Required):
◦ Page 1 — Population Overview: KPI cards (total patients, number of clusters, % in high-risk group). Cluster
size bar chart and risk-tier donut chart.
◦ Page 2 — Cluster Explorer: Interactive 2D scatter plot (PCA/UMAP/t-SNE projection). Clicking a cluster
shows its full clinical profile in a side panel (radar chart + feature table).
◦ Page 3 — Patient Risk Lookup: Enter patient ID or upload a single patient's lab values to receive their
cluster assignment, risk tier (Low/Medium/High), and a plain-language clinical summary.
◦ Page 4 — Feature Distributions: Side-by-side violin/box plots for any selected feature across all clusters,
to visually compare clinical differences.
◦ Page 5 — Batch Risk Scoring: Upload a CSV of new patients and download results with cluster labels, risk
tiers, and confidence scores (from GMM probabilities).
13. Clinical Safety Requirements:
◦ IMPORTANT: The dashboard must display a disclaimer on every page: 'This tool is for research and
decision-support only. It does not replace clinical judgment or diagnosis.'
◦ Confidence scores must be displayed alongside every prediction to communicate model uncertainty.
14. Deployment:
◦ Deploy on Streamlit Community Cloud, Hugging Face Spaces, or Render.
◦ Provide a [Link], a README with setup steps, and a short demo walkthrough video.
Deliverables
Deployed Clinical Dashboard: A live multi-page Streamlit or Dash app with all five required pages and the
mandatory disclaimer.
Deployment Documentation: Setup guide, architecture overview, and dashboard screenshots.

Milestone 5: Final Documentation and Presentation

Tasks
15. Final Report Must Include:
◦ Full methodology from data acquisition to clinical deployment.
◦ Clinical interpretation of each patient cluster with supporting statistics.
◦ Ethical considerations: patient data privacy (HIPAA/GDPR), algorithmic bias in clinical AI, and limitations
of unsupervised labels in medicine.
◦ Recommendations: how discovered groups can inform hospital resource planning, screening programs,
or personalized treatment protocols.
Deliverables
Final Project Report: Comprehensive clinical clustering report with ethical considerations and actionable
healthcare recommendations.

Milestones Summary
Milestone Key Deliverables

1. Data Collection & Medical EDA Medical EDA Report, Comorbidity Network, Cleaned
Dataset

2. Clinical Feature Engineering Feature Engineering Report, PCA & Projection

Visualizations

3. Clustering & Clinical Validation Clustering Report, Clinical Cluster Profiles, Model
Code

4. Clinical Decision-Support Dashboard Deployed Multi-Page Streamlit/Dash App

5. Final Documentation & Presentation Final Report.

 Alamein University Data Mining
Faculty of Computer Science Course code: AIE323
& Engineering
Bonus Project: Tweet Sentiment Analysis & Text Mining

Project Overview
This project focuses on analyzing tweets and short text posts to understand public opinion and sentiment trends.
Students will build a simple text mining pipeline that starts from raw social media data and ends with meaningful
insights and an interactive dashboard.
The goal is to explore how people feel about different topics using text data (positive, negative, or neutral sentiment)
and visualize patterns over time.

Milestone 1 Data Collection and Text Preprocessing

Objectives
• Collect tweet or short text data and prepare it for analysis.

Tasks
1. Data Collection:
◦ Collect tweets or social media posts using public datasets or APIs
◦ Focus on a specific topic (e.g., movies, products, events, or brands)
◦ Ensure data includes text and (if available) sentiment labels or ratings
2. Text Cleaning & Preprocessing:
◦ Remove URLs, hashtags, mentions, emojis, and special characters
◦ Convert text to lowercase
◦ Remove stop words (common words like “is”, “the”, “and”)
◦ Tokenize text into words

3. Text EDA:
◦ Generate word clouds per sentiment class, unigram/bigram frequency plots, and review length
distributions.
◦ Visualize class balance (positive, negative, neutral) and create a preprocessing impact comparison.

Milestone 2: Feature Extraction and Basic Models

Objectives
• Convert text into structured numerical form and build basic sentiment models.

Tasks
4. Feature Extraction:

o Convert text into numerical features using standard text representation methods
o Extract word frequency patterns and important terms
o Analyzing most influential words per sentiment class

5. Baseline Model Development:

◦ Train: Logistic Regression, Naive Bayes (Multinomial & Complement), Linear SVM, and Random Forest.
◦ Evaluate using Accuracy, Precision, Recall, F1-score (macro & weighted), and ROC-AUC.
 Alamein University Data Mining
Faculty of Computer Science Course code: AIE323
& Engineering
◦ Generate confusion matrices and per-class performance breakdowns.
6. Error Analysis:
◦ Manually review 30+ misclassified examples. Categorize error types (sarcasm, ambiguity, domain-
specific).
◦ Identify the most challenging linguistic patterns for baseline models.

Deliverables
Baseline Model Report: Comparative evaluation of all baseline models across both domains.
Error Analysis Report: Manual analysis of failure cases and identified linguistic challenges.

Milestone 3: Opinion Mining, Trend Analysis & Real-Time Dashboard

Objectives
• Extract structured opinions at scale and deploy a real-time sentiment monitoring dashboard with Streamlit or
Dash.

Tasks
7. Opinion Mining:
◦ Extract opinion phrases using dependency parsing: (opinion target, opinion expression, polarity) triplets.
◦ Build a structured opinion database from unstructured text at scale.
8. Trend Analysis:
◦ Plot sentiment trends over time (daily/weekly rolling averages).
◦ Detect sentiment shifts and correlate them with real-world events (product launches, news).
◦ Identify emerging negative topics using temporal keyword frequency analysis.
9. Streamlit / Dash Dashboard (Mandatory — All Features Required):
◦ Page 1 — Live Analyzer: Text input box; user types of any review and gets sentiment, confidence score,
and aspect breakdown in real time.
◦ Page 2 — Batch Processor: Upload a CSV file of reviews and download results with predicted labels and
ABSA output.
◦ Page 3 — Opinion Dashboard: Filterable aspect sentiment heatmap (product vs. sentiment), top
positive/negative opinion phrases.
◦ Page 4 — Trend Monitor: Time-series sentiment trend charts per topic/domain with anomaly highlights.
◦ Page 5 — Model Comparison: Side-by-side prediction from Logistic Regression vs. BERT for the same
input text.
◦ Arabic Toggle (Bonus): A language switcher that routes Arabic input to AraBERT and English to BERT.

Deliverables
Deployed Real-Time Dashboard: A fully functional multi-page Streamlit or Dash app with all required pages.
Opinion Mining Module: Code for structured (target, expression, polarity) triplet extraction.
Trend Analysis Report: Documented sentiment trends with event correlations.

Milestone 5: Final Documentation and Presentation

Tasks
10. Final Report Must Include:
◦ End-to-end methodology: data collection, preprocessing, baseline models, and dashboard.
 Alamein University Data Mining
Faculty of Computer Science Course code: AIE323
& Engineering
Deliverables
Final Project Report: Comprehensive multi-domain NLP project report with ethical considerations.

Milestones Summary
Milestone Key Deliverables
1. Multi-Domain Data & NLP Pipeline Multi-Domain Dataset, NLP Pipeline, Text EDA Report

2. Feature Engineering & Baseline, Models Baseline Model Report, Error Analysis Report

3. Opinion Mining & Real-Time Dashboard Deployed Multi-Page Streamlit/Dash App

4. Final Documentation & Presentation Final Report