This repository serves as a collection of my work and learning in machine learning while my internship in Cellual-Technologies, including algorithm explanations, data preprocessing workflows, and two projects.
Note: The supervised learning section is incorrectly numbered as "II" instead of "I". This should be fixed.
Before training any machine learning model, we go through Exploratory Data Analysis (EDA) and Data Preprocessing.
These steps ensure that the dataset is clean, consistent, and ready for modeling.
EDA helps understand the dataset’s structure, patterns, and potential issues.
- Understanding the data – Checking data types, dimensions, and sample values.
- Statistical summary – Using
describe()to find mean, median, min, max, etc. - Missing values – Identifying and deciding how to handle NaN values.
- Data distribution – Plotting histograms, KDE plots, and boxplots.
- Outlier detection – Using visualization and statistical methods like IQR.
- Correlation analysis – Finding relationships between variables with heatmaps.
Once we understand the data, preprocessing ensures it’s ready for algorithms.
- Handling missing values – Imputation with mean/median/mode or removal.
- Encoding categorical variables – One-hot encoding or label encoding.
- Feature scaling – Normalization or Standardization for numerical features.
- Feature engineering – Creating new features from existing ones.
- Splitting data – Training/testing (and validation) sets to evaluate models.
flowchart TD
A[Data Collection] --> B[Exploratory Data Analysis]
B --> C[Data Preprocessing]
C --> D[Model Selection]
D --> E[Training]
E --> F[Evaluation]
F --> G[Deployment]
Below is a categorized explanation of various algorithms.
Algorithms that find patterns with labeled data.
- Type: Regression
- Use case: Predicting continuous values.
- Concept: Fits a straight line that minimizes the difference between predicted and actual values (using least squares method).
- Key points:
- Assumes a linear relationship between variables.
- Sensitive to outliers.
- Example: Predicting house prices.
- Type: Regression
- Use case: Modeling non-linear relationships.
- Concept: Extends linear regression by adding polynomial terms (x², x³, …).
- Key points:
- Fits a curve instead of a straight line.
- Risk of overfitting with high polynomial degree.
- Type: Classification
- Use case: Predicting binary or multi-class categories.
- Concept: Uses a sigmoid function to output probabilities for class membership.
- Key points:
- Despite its name, it’s a classification algorithm.
- Example: Spam detection.
- Type: Classification/Regression
- Use case: Classifying data points based on their closest neighbors.
- Concept: Looks at the “k” nearest data points and assigns the majority class (classification) or average (regression).
- Key points:
- Simple, non-parametric method.
- Computationally expensive for large datasets.
- Type: Classification/Regression
- Use case: Separating data into distinct classes with the widest possible margin.
- Concept: Finds an optimal hyperplane that maximizes the margin between classes.
- Key points:
- Works well with high-dimensional data.
- Can use kernels for non-linear separation.
- Type: Classification
- Use case: Text classification, spam filtering.
- Concept: Based on Bayes’ theorem with the assumption of feature independence.
- Key points:
- Fast and efficient.
- Works well with high-dimensional data (e.g., text).
- Type: Classification/Regression
- Use case: Predicting classes or values by splitting data into branches.
- Concept: Divides data based on feature values until reaching a decision.
- Key points:
- Easy to interpret.
- Can overfit without pruning.
- Type: Classification/Regression
- Use case: More robust version of Decision Tree.
- Concept: Combines multiple decision trees (ensemble) and averages results.
- Key points:
- Reduces overfitting.
- Works well for a wide range of problems.
Algorithms that find patterns without labeled data.
- Type: Clustering
- Use case: Grouping similar data points.
- Concept: Partitions data into “k” clusters by minimizing distances within clusters.
- Key points:
- Requires specifying “k” in advance.
- Sensitive to initial centroids.
- Type: Clustering
- Use case: Creating a hierarchy of clusters.
- Concept: Builds nested clusters using a tree-like diagram (dendrogram).
- Key points:
- No need to predefine number of clusters.
- Computationally intensive for large datasets.
- Type: Clustering
- Use case: Identifying clusters of arbitrary shape.
- Concept: Groups together points that are closely packed and marks outliers.
- Key points:
- No need to specify number of clusters.
- Handles noise well.
- Type: Feature reduction
- Use case: Reducing high-dimensional data while retaining variance.
- Concept: Transforms features into new uncorrelated variables (principal components).
- Key points:
- Speeds up computation.
- Useful for visualization of complex datasets.
Evaluating machine learning models is essential to understand how well they generalize to unseen data.
Below are common evaluation metrics grouped by problem type.
Measures the average magnitude of errors without considering their direction.
MAE = (1/n) * Σ |yᵢ - ŷᵢ|
- Pros: Easy to interpret, less sensitive to outliers than MSE.
- Cons: Does not penalize large errors as strongly.
Measures the average of squared differences between actual and predicted values.
MSE = (1/n) * Σ (yᵢ - ŷᵢ)²
- Pros: Penalizes large errors more than MAE.
- Cons: Sensitive to outliers.
Square root of MSE, bringing it back to the same units as the target variable.
RMSE = √( (1/n) * Σ (yᵢ - ŷᵢ)² )
- Pros: More interpretable than MSE.
- Cons: Same sensitivity to outliers as MSE.
Measures the proportion of variance in the dependent variable explained by the model.
R² = 1 - [ Σ (yᵢ - ŷᵢ)² / Σ (yᵢ - ȳ)² ]
- Range: 0 to 1 (Higher is better, negative means worse than a horizontal line).
- Pros: Gives a percentage interpretation.
- Cons: Can be misleading for non-linear models.
The proportion of correct predictions out of total predictions.
Accuracy = (TP + TN) / (TP + TN + FP + FN)
Measures the percentage of positive predictions that are actually correct.
Precision = TP / (TP + FP)
Measures the percentage of actual positives correctly identified.
Recall = TP / (TP + FN)
Harmonic mean of Precision and Recall.
F1 = 2 * (Precision * Recall) / (Precision + Recall)
- ROC Curve: Plots True Positive Rate vs. False Positive Rate at different thresholds.
- AUC: Measures the overall ability of the model to discriminate between classes.
- Range: 0 to 1 (Higher is better).
Measures how similar an object is to its own cluster compared to other clusters.
s = (b - a) / max(a, b)
Where:
a= mean intra-cluster distanceb= mean nearest-cluster distance
Measures the average similarity between each cluster and its most similar cluster.
DB = (1/n) * Σ maxⱼ≠ᵢ [ (σᵢ + σⱼ) / d(cᵢ, cⱼ) ]
Where:
σ= average distance between points in a cluster and the cluster centroidd(cᵢ, cⱼ)= distance between cluster centroids
Lower DB index means better clustering.
✅ Tip: Always choose the metric based on the problem type and business goal. For example:
- Regression → MAE, RMSE, R²
- Classification → F1, Precision-Recall, ROC-AUC
- Clustering → Silhouette, Davies–Bouldin
This repository combines theory and practice, providing algorithm explanations and real project implementations. It can be used as a reference for machine learning studies and practical applications.