https://github.com/kalana99/190530h_ml_mini_project

A Mini project to critically evaluate various classifier models along with feature engineering techniques.
https://github.com/kalana99/190530h_ml_mini_project

classification-algorithm feature-engineering feature-extraction feature-selection hyperparameter-tuning kaggle-competition machine-learning pca

Last synced: about 1 month ago
JSON representation

A Mini project to critically evaluate various classifier models along with feature engineering techniques.

• Host: GitHub
• URL: https://github.com/kalana99/190530h_ml_mini_project
• Owner: Kalana99
• Created: 2023-09-23T17:12:52.000Z (about 1 year ago)
• Default Branch: main
• Last Pushed: 2023-10-16T11:24:08.000Z (about 1 year ago)
• Last Synced: 2023-10-17T03:23:19.270Z (about 1 year ago)
• Topics: classification-algorithm, feature-engineering, feature-extraction, feature-selection, hyperparameter-tuning, kaggle-competition, machine-learning, pca
• Language: Jupyter Notebook
• Homepage:
• Size: 36.1 KB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# 190530H_ML_Mini_Project

## Overview

Dataset: AudioMNIST is the dataset used to create the features. Check this link for further

details about the dataset Link.

This project has two phases:

1. Individual task - Classification model development and Kaggle competition for each assigned layer.

2. Group submission - 6-page research paper

The above notebooks provide classification models for **Layer 09** and **Layer 10**.

## Data Pre-processing and Feature Engineering Steps

- Load data sets from the Google Drive.

- Identify labels with missing values.

- Drop the entries with missing values and Scale for each label to handle outliers.

- Create dictionaries for each label using Train, Valid, and Test data sets.

- Apply feature selection using **SelectKBest** and **f_classif**.

- Conduct PCA analysis.

## Training the Final Model

- **SVM**, **K-NN**, and **Random Forest** are considered possible candidates for the classifier model. Each of these models is tested with pre-processed data. Based on the results, SVM stands out as the most suitable model.

- **Hyper-parameter Tuning** is conducted using a **Random Grid Search** for an SVM instance. The resulting best estimator is used to predict the labels using the Test dataset.

- The result is written into a CSV file in the required format.

*The above operation sequence is carried out for each label separately by changing the label in the **“Assigning Labels”** cell.*

```Python

# @title **Assigning Label**

train_label = 'label_X'

x_train_df = x_train[train_label].copy()

y_train_df = y_train[train_label].copy()

x_valid_df = x_valid[train_label].copy()

y_valid_df = y_valid[train_label].copy()

x_test_df = x_test[train_label].copy()

```

Replace **X** with **1, 2, 3** or **4** based on the label that needs to be predicted