Maching Learning Algorithms From Scratch

This repository contains implementations of various machine learning algorithms I have done from scratch.

• Linear Discriminant Analysis (LDA)

  • Reduce the dimensionality of features space from 30-D to 1-D using Linear Discriminant Analysis algorithm with the goal of keeping the class information intact.
  • Dataset used: Breast Cancer Wisconsin Dataset

• Deicsion Tree

  • Predict the class of iris plant using avilable features, a multi-class classification task.
  • Dataset used: Iris

• K-Means Clustering

  • Segment customers to aid the marketing team and plan streategy accordingly.
  • Dataset used: Mall Customer Segmentation Data

• K-Nearest Neighbours (KNN)

  • Predict whether a patient has diabetes based on diagnostic measurements.
  • Dataset used: Diabetes Dataset

• Linear Regression

  • Predict the quality of wine with the assistance of their physiochemical properties.
  • Dataset used: Red Wine Quality

• Logistic Regression

  • Predict whether the cancer type is Malignant or Benign using Logistic Regression algorithm, and achieved 98% accuracy and 0.97 F1-score and on test set.
  • Dataset used: Breast Cancer Wisconsin Dataset

• Moving Averages

  • Apply Simple Moving Average, Commulative Moving Average, Weighted Moving Average, and Exponential Moving Average techniques on Air Quality Index
  • Dataset used: Air Quality Data Set

• Naive Bayes

  • Classify the cancer type Malignant or Benign using Gaussian Naive Bayes algorithm, and achieved 90% accuracy on test set.
  • Dataset used: Breast Cancer Wisconsin Dataset

• Principal Component Analysis (PCA)

  • Reduce the dimenionsality of face images by approximately 250x using Principal Component Analysis (PCA) and regenerate them using basis vectors.
  • Dataset used: The Database of Faces (AT&T)

• Perceptron

  • Demonstrate perceptron failure on non-seprable data.
  • Additionally, predict whether a patient has diabetes based on diagnostic measurements.
    • Dataset: Diabetes Dataset

• Simple Neural Network

  • Designed a simple neural network (with forward and backward path) consists of an input layer, an output layer and a hidden layer.
  • Mimicking the behaviour of XOR gate.

• Support Vector Machine

  • Performed Binary classification to predict cancer types Benign or Malignant achieving up to 97% accuracy and 0.95 F1-Score on the test set.
  • Dataset: Breast Cancer Wisconsin Dataset