AI_Portfolio

This repository contains four Jupyter notebooks showcasing applied artificial intelligence projects developed in Python using real-world datasets. The work includes multiple data modalities tabular (task 1), image (task 2), audio (task 3), and text data (task 4) and compares classical machine learning models with deep learning models.

Project Overview

The portfolio consists of the following AI tasks:

1. Tabular Data – Income Prediction

• Dataset: Adult Income (UCI Repository)
• Task: Binary classification
• Techniques: Logistic Regression, Decision Trees, Random Forests, Gradient Boosting, Neural Networks
• Evaluation: Accuracy, Precision, Recall, F1-score, ROC-AUC
• Explainability: Feature importance and SHAP analysis

2. Image Data – Crop Disease Classification

• Dataset: PlantVillage
• Task: Multi-class classification
• Techniques: Classical ML with feature extraction, Convolutional Neural Networks (CNNs)
• Evaluation: Accuracy, Precision, Recall, F1-score
• Explainability: Grad-CAM visualisations

3. Audio Data – Urban Sound Classification

• Dataset: UrbanSound8K
• Task: Multi-class classification
• Techniques: Feature extraction (MFCCs), classical ML models, CNNs on spectrograms
• Explainability: SHAP / LIME analysis
• Additional: Unsupervised clustering for comparison

4. Text Data – Fake News Detection

• Dataset: Fake and Real News (Kaggle)
• Task: Binary classification
• Techniques: TF-IDF with classical ML models, fine-tuned Transformer models
• Evaluation: Precision, Recall, F1-score, ROC-AUC
• Explainability: Word importance and attention analysis

---

Key Topics Covered

• Classical Machine Learning
• Deep Learning (Neural Networks, CNNs)
• Model Evaluation and Visualisation
• Explainability (SHAP, LIME, Grad-CAM)
• Fairness and Ethical Considerations in AI
• Real-world dataset handling

---

Tools and Technologies

• Python
• Jupyter Notebooks
• NumPy, pandas, scikit-learn
• TensorFlow / PyTorch
• Matplotlib / Seaborn

---

Notes

All results, visualisations, and outputs are included directly within the notebooks. Each notebook also contains a short written summary discussing performance, interpretability, and limitations.

This repository was developed as part of a BEng Computer Science Artificial Intelligence module and demonstrates practical, real-world application of modern AI techniques.