This repository is a comprehensive collection of experiments and projects developed during a specialized Machine Learning course. It systematically covers fundamental concepts and advanced applications across both Supervised Learning and Unsupervised Learning paradigms.
The structure is divided into three main modules (C1, C2, and C3), progressing from basic prediction models to deep data evaluation and complex clustering tasks, serving as a robust professional portfolio.
- Focus: Introduction to building and evaluating basic supervised prediction models. This section establishes the fundamental workflow: data preparation, model training, and performance measurement.
- Key Techniques Covered:
- Regression Modeling: Building models to estimate continuous target variables (e.g., estimating rental prices based on location and features).
- Classification: Implementing and evaluating initial classification models.
- Model Comparison: Early assessment of model performance and the risks of overfitting (e.g., comparing models on small datasets where 100% accuracy may be misleading).
This module intensifies the focus on the end-to-end Machine Learning pipeline, emphasizing data cleaning, feature engineering, and rigorous model evaluation in real-world scenarios.
- Focus: Developing robust predictive models for high-stakes business decisions (e.g., housing investment), with a strong emphasis on data quality, error analysis, and actionable business recommendations.
- Key Techniques Covered:
- Data Preprocessing: Thorough data cleaning, format correction, handling of Outliers and Missing Values.
- Regression Algorithms: Implementing and comparing advanced regression models, specifically leveraging RandomForestRegressor for its accuracy and resilience to extreme values.
- Feature Engineering & Selection: Selection of the most relevant variables to simplify models and improve consistency (e.g., analyzing the correlation between stratum and administration price in housing data).
- Model Evaluation: Detailed analysis of prediction errors (residuals), ensuring errors are distributed normally and without bias.
- Business Application: Translating model results into clear, justified business recommendations (e.g., recommending a specific model for property price prediction for investment planning).
This module shifts the focus to Unsupervised Learning, utilizing clustering methods to find hidden structures and patterns in complex, high-dimensional datasets, including image data.
- Focus: Applying K-Means Clustering to various data distributions and complex problems like image grouping, and mastering the techniques for determining the optimal number of clusters.
-
Key Techniques Covered:
- K-Means Algorithm: Implementation and application of the algorithm on varied synthetic and real-world datasets.
-
Optimal Cluster Determination: Using both the Elbow Method (Elbow Curve) and the Silhouette Coefficient to rigorously determine the most appropriate number of groups (
$k$ ) for a given dataset. - High-Dimensional Data Analysis (Images): Applying clustering to high-dimensional image data (e.g., face classification or general image grouping), including addressing the challenges of high-dimensionality (49152 dimensions).
- Cluster Interpretation: Visualizing cluster centroids (average images) to understand the predominant characteristics of each group and interpreting the limitations of clustering in high-dimensional space.
The projects are developed using Python and standard scientific computing libraries.
| Category | Libraries / Tools |
|---|---|
| Language | Python |
| Data Manipulation | pandas, numpy |
| Visualization | matplotlib, seaborn |
| Supervised Learning | scikit-learn (sklearn), RandomForestRegressor, KNeighborsClassifier, LogisticRegression |
| Unsupervised Learning | KMeans, PCA (Dimensionality Reduction) |
| Evaluation Metrics | MSE, RMSE, Confusion Matrix, Accuracy, Precision, Recall, Elbow Method, Silhouette Coefficient |