Machine Learning Projects

A curated portfolio of end-to-end machine learning systems spanning ranking, recommendation, robustness under distribution shift, and quality-of-experience prediction.
Each project is structured to demonstrate strong ML fundamentals, evaluation rigor, and production-shaped engineering practices (reproducible pipelines, modular code, artifact outputs).

---

What’s in this repository

This repository currently contains the following projects:

Project	Domain	Core ML Task	Key Metrics
Search Ranking System for Streaming Content	Search & Discovery	Learning-to-Rank	NDCG@K, MRR@K, Precision/Recall@K
personalized-recsys	Personalization	Recommendation & Ranking	MAP@K, NDCG@K, Recall@K (project-specific)
robustness-under-shift-qoe	Reliability	Robustness under Distribution Shift	Calibration, performance under shift, error analysis
streaming-quality-prediction	Media / QoE	Prediction & Monitoring	RMSE/MAE (or classification metrics), trend/anomaly analysis

Repo folders: Search Ranking System for Streaming Content/, personalized-recsys/, robustness-under-shift-qoe/, streaming-quality-prediction/.

---

Technical Focus Areas

• Learning-to-Rank (LTR): query/session grouping, pairwise objectives, offline ranking metrics
• Recommender Systems: personalization signals, candidate ranking, top-K evaluation
• Robust ML: distribution shift evaluation, stability analysis, diagnostic reporting
• Quality-of-Experience Modeling: feature engineering, prediction, monitoring-style evaluation
• Engineering Practices: reproducible runs, deterministic seeds, modular pipeline stages, artifact outputs

---

Tech Stack (common)

• Language: Python
• Core Libraries: NumPy, pandas, scikit-learn
• Modeling: XGBoost / tree models / classical ML (project-dependent)
• Evaluation: ranking metrics (NDCG/MRR/MAP), regression/classification metrics
• Artifacts: parquet/CSV outputs, serialized models, metric JSON reports

---

Projects

1) Search Ranking System for Streaming Content

Goal: Build a learning-to-rank pipeline that ranks content results for a given query/session using a multi-stage workflow: data → sessions → features → ranker training → offline evaluation → inference.

What it demonstrates

• Query/session-aware ranking (grouped training)
• Feature engineering spanning relevance + personalization + popularity signals
• Pairwise ranking model training and evaluation with ranking metrics

Typical outputs

• Sessionized training data (e.g., sessions.parquet)
• Feature table (e.g., features.parquet)
• Model artifact (e.g., ranker.joblib)
• Offline metrics report (e.g., metrics.json)

Metrics

• NDCG@K, MRR@K, Precision@K, Recall@K

---

2) personalized-recsys

Goal: Build a personalized recommendation workflow that ranks items per user using interaction signals and content metadata.

What it demonstrates

• Personalization signal design (user affinity / item popularity / content similarity)
• Top-K evaluation that mirrors recommendation workloads
• Modular separation of retrieval vs ranking (if implemented in the project)

Metrics (examples)

• NDCG@K, MAP@K, Recall@K, HitRate@K (based on the project implementation)

---

3) robustness-under-shift-qoe

Goal: Study model performance under distribution shift and build diagnostics that explain failure modes.

What it demonstrates

• Shift-aware evaluation (train vs test mismatch)
• Robustness analysis (performance stratified by segments)
• Error analysis and metric reporting for reliability-focused ML

Metrics (examples)

• Performance deltas under shift (e.g., ΔAUC/ΔRMSE), calibration error, slice-based analysis

---

4) streaming-quality-prediction

Goal: Predict streaming quality/QoE signals using engineered features and evaluate model quality for monitoring use-cases.

What it demonstrates

• Feature engineering for operational prediction tasks
• Regression/classification modeling (depending on labels)
• Reporting that aligns with monitoring/decision-support workflows

Metrics (examples)

• RMSE/MAE for regression; Precision/Recall/F1/AUC for classification (based on the project implementation)

---

Getting Started

1) Clone the repository

git clone https://github.com/Uttam-38/Machine_Learning_Projects-.git cd Machine_Learning_Projects-

2) Create an environment

python -m venv .venv

macOS/Linux : source .venv/bin/activate

Windows : venv\Scripts\activate

3) Install dependencies

pip install -U pip pip install numpy pandas scikit-learn matplotlib xgboost