README.md #ML Classification Tool

Built to understand, not just to use.

Most ML courses teach students to call sklearn.fit() without explaining what happens inside. This notebook solves that — it implements KNN, Naive Bayes, and Decision Tree from scratch, so every step of the learning process is visible and traceable.

---

The Problem

When students use scikit-learn, they treat algorithms as black boxes:

from sklearn.neighbors import KNeighborsClassifier
model = KNeighborsClassifier()
model.fit(X_train, y_train)   # What actually happens here?

This works, but it builds no real understanding. A student who only uses sklearn cannot explain why KNN fails on high-dimensional data, or what Naive Bayes assumes about feature distributions.

The Solution

This notebook replaces the black box with transparent code. Every algorithm is written using only Python built-ins and NumPy — no sklearn for training. A student running this notebook will see exactly how:

• KNN measures distance and picks neighbors
• Naive Bayes computes probabilities per class
• A Decision Tree chooses split points using Gini impurity

The same pipeline works on any CSV dataset, making it reusable across different courses and experiments.

---

Full Pipeline

Step	What Happens
Load	Reads CSV from local path or Google Drive link
Clean	Removes duplicates, fills missing values (mean / mode)
Encode	Maps text columns to integers using dictionaries
Split & Scale	80/20 split, standardization using train-set stats only
Train	Three classifiers built manually with NumPy
Evaluate	Accuracy, confusion matrix, side-by-side comparison chart

---

Algorithms — Implemented From Scratch

K-Nearest Neighbors Computes Euclidean distance between the test row and every training row, then predicts by majority vote among the k closest neighbors. Uses (distance, label) tuples — no external sorting library.

Gaussian Naive Bayes Estimates the mean and standard deviation of each feature per class during training. At prediction time, picks the class with the highest log-posterior probability using the Gaussian probability density formula.

Decision Tree Builds a recursive binary tree up to depth 3. At each node, it searches every feature and threshold to find the split that minimizes Gini impurity. Returns a nested dictionary — no objects or classes required.

---

Tech Stack

• Python 3
• NumPy — all algorithm math
• Matplotlib + Seaborn — visualization only
• scikit-learn confusion_matrix — evaluation only
• Google Colab compatible

---

How to Run

Option 1 — Google Colab (recommended)

Open the notebook in Colab and run all cells in order.

Option 2 — Local Jupyter

pip install numpy matplotlib seaborn scikit-learn
jupyter notebook fainalML_v6_simple_chapters.ipynb

Sample Dataset

A ready-to-use dataset is available at:

https://drive.google.com/file/d/16AMnMYPwkpWCCO6cZEWzXfUC_fyvnonO/view?usp=sharing

Paste this link when the notebook prompts for input.

---

Who Is This For

• Students learning ML who want to understand algorithms beyond the sklearn API
• Anyone who wants a clean, dependency-light classification baseline they can read and modify
• Courses that require implementing algorithms without using pre-built models

---

Implementation Notes

• No data leakage — scaling parameters are computed on the training set only, then applied to the test set
• No sklearn for training — all three models are written from first principles
• Encoding transparency — every text-to-integer mapping is stored and printed so the user can trace predictions back to original labels
• Robust CSV handling — supports UTF-8 and Latin-1 encodings, empty rows, and inconsistent column widths