Advanced Data Analytics

This repo contains my work for an Advanced Data Analytics course. It is organized into three assignments, each with a classification and a regression notebook. The datasets come from UCI Machine Learning Repository and OpenML.

---

First Assignment — EDA and Modeling

The focus here was on exploratory data analysis using profiling tools like ydata_profiling, sweetviz, and autoviz, then building some baseline models.

Classification — Predictive Maintenance dataset (AI4I 2020, UCI)

• 10,000 rows, synthetic industrial data simulating machine failures
• Target: whether a machine failed or not (binary)
• Explored failure types: heat dissipation failure, power failure, overstrain failure, tool wear failure
• Used Logistic Regression, Random Forest, and Gradient Boosting
• Handled class imbalance with sample weights and stratified cross-validation

Regression — Iranian Telecom Customer Churn dataset (UCI)

• 3,150 customers, 13 features
• The dataset is originally for churn classification, but I used it for regression with Customer Value as the target
• Used Linear Regression, Ridge, Huber, and Random Forest regressors
• Preprocessing with RobustScaler inside a pipeline

---

Second Assignment — Data Quality and Sampling

This assignment dealt with messier problems: class imbalance, noise, overlapping classes, and outliers. The goal was to try different data preparation strategies and compare results.

Classification — Codon Usage dataset (UCI)

• Genomic data mapping codon usage frequencies to biological kingdoms (bacteria, virus, plant, mammal, etc.)
• Large dataset with strong class imbalance
• Tried sampling methods: RandomOverSampler, SMOTENC, ADASYN, TomekLinks, RandomUnderSampler
• Feature selection with mutual information and VIF to reduce multicollinearity
• Models: Random Forest, KNN inside an imbalanced pipeline

Regression — Steel Industry Energy Consumption dataset (UCI)

• Electricity usage data from a steel plant in South Korea
• Target: Usage_kWh
• Worked on outlier detection and removal using IsolationForest, RANSAC, and EllipticEnvelope
• Used CTGAN for synthetic data generation
• Models: Random Forest Regressor

---

Third Assignment — Feature Engineering

The focus shifted to feature selection, multicollinearity, dimensionality reduction, and building new features.

Classification — EEG Eye State dataset (OpenML)

• EEG recordings from a single 117-second session using an Emotiv headset
• Target: whether eyes are open or closed (binary)
• Time-series structure, loaded from ARFF format using scipy
• Feature selection with mutual information and phik correlation
• Applied PCA and created time-based features
• Used TimeSeriesSplit for cross-validation

Regression — PGCB Hourly Generation dataset (UCI)

• Hourly electricity generation, demand, and loadshedding data from Bangladesh's national grid
• Target: demand_mw
• Feature selection using mutual information and F-test
• Created interaction features and cluster-based profiles
• Checked redundancy with PCA and VIF
• Models: Random Forest Regressor

---

Structure

First-Assignment/
    Classification.ipynb     # Predictive Maintenance (machine failure)
    Regression.ipynb         # Customer Value prediction
    Datasets/

Second_Assignment/
    classification.ipynb     # Codon Usage (biological kingdoms)
    Regression.ipynb         # Steel Industry energy consumption
    datasets/

Third_assignment/
    Eeg_Eye_state_classification.ipynb   # EEG eye open/closed
    PGCB_Regession.ipynb                 # Bangladesh electricity demand
    Datasets/

---

Libraries Used

• pandas, numpy, matplotlib, seaborn
• ydata_profiling, sweetviz, autoviz
• scikit-learn, imbalanced-learn
• phik, statsmodels
• ctgan (synthetic data generation)
• mne, scipy (EEG data handling)