Welcome to my comprehensive Machine Learning repository! This project serves as a structured, hands-on guide covering essential machine learning algorithms, critical data preprocessing techniques, and robust feature engineering workflows using Python and scikit-learn.
The repository is organized into distinct categories, tracking the journey from raw data exploration to advanced predictive modeling.
Before modeling, understanding data distributions and behaviors is essential:
understanding-your-data-descriptive-stats— Initial data profiling, shape analysis, and basic metrics.univariate-analysis— Exploring individual variables independently.bivariate-analysis— Analyzing relationships between pairs of variables.pandas-profiling— Automated exploratory data analysis reports.
Real-world data lives in various formats and across networks:
Api-to-dataframe— Fetching live data via APIs and converting it into structured DataFrames.Pandas-dataframe-using-web-scraping— Gathering tabular data directly from web scraping workflows.Working with csv files— Standard file operations and optimization tricks.Working-with-json-and-sql— Parsing unstructured nested JSON data and executing relational SQL queries.
Data preparation determines model quality. This section handles missing data, encoding, scaling, and distribution transformations:
- Missing Value Imputation:
complete-case-analysis— Handling missing data through listwise deletion.imputing-numerical-data— Statistical measures (Mean/Median) for numeric gaps.handling-missing-categorical-data— Mode and frequent category replacements.missing-indicator— Flagging missing values explicitly to retain variance information.knn-imputer— K-Nearest Neighbors multivariate imputation.iterative-imputer— Advanced multivariate imputation by modeling each feature as a function of others.
- Categorical Encoding:
ordinal-encoding— Converting ordered categories into rank integers.one-hot-encoding— Transforming nominal fields into binary columns.
- Numerical Scaling & Transformation:
standardization— Centering distributions around zero with unit variance.normalization— Scaling ranges directly between a fixed bound (0 and 1).binning-and-binarization— Discretizing continuous data into categorical buckets or binary thresholds.function-transformer— Applying custom log, exponential, or reciprocal mathematical edits.power-transformer— Stabilizing variance using Box-Cox or Yeo-Johnson transforms to yield normal distributions.
- Complex Feature Interactions:
handling-mixed-variables— Parsing fields containing both numeric strings and text markers.handling-date-and-time— Extracting operational features like day, month, hour, and weekend statuses.feature-construction-and-feature-splitting— Engineering completely new indicators or splitting text composites.
column-transformer— Applying entirely different preprocessing pipelines to distinct columns simultaneously.sklearn-pipelines— Chaining preprocessing blocks cleanly with estimator classes for leakage-free cross-validation.
Methods for detecting, trimming, or capping distribution extremes:
outlier-detection-using-percentiles— Quantile-based filtering strategies.outlier-removal-using-zscore— Identifying data extremes based on standard deviation distances.outlier-removal-using-iqr-method— Leveraging the Interquartile Range to build non-parametric box boundaries.
simple-linear-regression— Single-variable continuous targets.multiple-linear-regression— Multi-variable linear combinations.polynomial-regression— Engineering non-linear feature degrees for curved trends.gradient-descent&types-of-gradient-descent— Deep dive into optimization routines (Batch, Stochastic, and Mini-Batch).- Regularized Models:
lasso-regression— L1 Regularization for embedded feature selection.lasso-regression— L2 Regularization to combat multicollinearity.elasticnet-regression— Combining L1 and L2 penalties linearly.regularized-linear-models— Comprehensive comparative suite.
logistic-regression&logistic-regression-contd— Foundational probabilistic binary and multi-class classification.
adaboost— Adaptive boosting algorithms focusing sequentially on hard-to-classify samples.random-forest— Bootstrap aggregating (bagging) over decorative decision tree branches.pca— Principal Component Analysis for orthogonal dimensionality reduction and noise extraction.
regression-metrics— Measuring continuous error (MAE, MSE, RMSE, R²).classification-metrics— Gauging discrete performance (Accuracy, Precision, Recall, F1-Score, ROC-AUC).
Make sure you have Python 3.8+ installed on your computer.
Clone this repository and enter the directory:
git clone https://github.com
cd Machine-LearningInstall the core dependencies:
pip install numpy pandas scikit-learn matplotlib seaborn jupyterLaunch Jupyter Notebook or JupyterLab to interactively explore and run any of the topic directories:
jupyter notebook