applied-ai-models

Applied AI Models Portfolio (ML/DL • NLP • CV • Recommenders)

Job-oriented portfolio showcasing practical implementations of Machine Learning & Deep Learning tasks with clear evaluation, model comparison, and reproducible experiments.
Includes classical ML pipelines, clustering + PCA, recommender systems, and deep learning for Computer Vision and NLP.

---

Project tags (keywords)

python machine-learning deep-learning ai data-science
numpy pandas matplotlib seaborn
scikit-learn svm random-forest feature-engineering
kmeans pca clustering
recommender-systems surprise svd svdpp nmf
tensorflow keras cnn transfer-learning vgg16 fashion-mnist
nlp imdb rnn lstm bilstm text-summarization nltk spacy

---

Why this repo

This repository demonstrates:

• End-to-end ML workflow: data loading → preprocessing → training → evaluation → conclusions
• Model comparison beyond accuracy: classification_report, confusion matrices, learning curves, cross-validation where applicable
• Reproducibility: fixed seeds, consistent preprocessing, modular helper code
• Engineering mindset: structured repo (notebooks/ + src/), clean code style, readable experiments

---

Highlights

Classical ML: Human Activity Recognition

• 05_human_activity_recognition_svm_rf.ipynb
  • Time-domain feature engineering
  • SVM vs RandomForest
  • classification_report + confusion matrix comparison

Clustering: KMeans + PCA visualization

• 06_kmeans_clustering_pca_visualization.ipynb
  • Elbow method
  • PCA to 2D for high-dimensional dataset
  • Cluster visualization & interpretation

Recommender Systems: Matrix Factorization

• 07_recommender_systems_matrix_factorization.ipynb
  • Surprise: SVD / SVD++ / NMF
  • Cross-validation and best model selection (RMSE/MAE)
  • Simple Top-N recommendation demo

Computer Vision: Fashion-MNIST (Keras)

• 08_fashion_mnist_mlp_baseline.ipynb — Dense MLP baseline (target ≥ ~0.91)
• 09_fashion_mnist_cnn_architecture_comparison.ipynb — CNN architecture + comparison to MLP
• 10_fashion_mnist_transfer_learning_vgg16.ipynb — Transfer learning VGG16 (feature extraction + fine-tuning)

NLP: IMDB Sentiment + Summarization (NLTK & spaCy)

• 11_imdb_sentiment_rnn_lstm_bilstm.ipynb
  • SimpleRNN vs LSTM vs BiLSTM (+ optional deep variant)
  • classification report + training diagnostics
• 12_text_summarization_nltk_spacy.ipynb
  • Extractive summarization using NLTK and spaCy
  • Sentence scoring with word frequencies + heapq.nlargest

---

Tech stack

Core: Python 3.10+, NumPy, Pandas
Visualization: Matplotlib, Seaborn
ML: scikit-learn
Deep Learning: TensorFlow / Keras
NLP: NLTK, spaCy
Recommenders: scikit-surprise

Links:

• NumPy: https://numpy.org/
• Pandas: https://pandas.pydata.org/
• scikit-learn: https://scikit-learn.org/stable/
• TensorFlow: https://www.tensorflow.org/
• Keras: https://keras.io/
• NLTK: https://www.nltk.org/
• spaCy: https://spacy.io/
• Surprise: https://surpriselib.com/

---

Setup

Option A — Local (pip)

python -m venv .venv
# Windows: .venv\Scripts\activate
# macOS/Linux: source .venv/bin/activate
pip install -r requirements.txt
jupyter notebook