Machine Learning Projects

📌 Introduction

This repository is dedicated to machine learning projects with the primary goal of learning and practicing ML concepts.

The projects included here are meant to:

• Explore different areas of machine learning.
• Practice the full workflow: data preprocessing, exploratory analysis, model development, and evaluation.
• Build both theoretical understanding and practical coding skills.

🎯 Purpose

The main purpose of this repository is to serve as a personal learning space while also providing others with examples and references for their own ML journey.

Each project is a step toward improving skills in data science and machine learning through hands-on implementation.

---

📂 Project 1: Sales Prediction

📖 Task Description
Given a product and its corresponding details, the task is to predict the amount of sales based on the available features.

• Dataset Size: ~8k rows
• Training Dataset Link: Google Sheets Dataset

📊 Dataset Columns

• Item_Identifier: Unique identity number for a product
• Item_Weight: Weight of the product
• Item_Fat_Content: Fat content (Low Fat / Regular)
• Item_Visibility: Percentage of store display allocated to the product
• Item_Type: Category of the product
• Item_MRP: Maximum Retail Price (list price)
• Outlet_Identifier: Unique store ID
• Outlet_Establishment_Year: Year the store was established
• Outlet_Size: Store size in terms of ground area
• Outlet_Location_Type: Type of city in which the store is located
• Outlet_Type: Type of outlet (grocery store or supermarket)
• Item_Outlet_Sales: Target variable – product sales in the store

🤖 Recommended Approaches & Models

• Data Preprocessing: Handle missing values, encode categorical features, and scale numeric values.
• Exploratory Analysis: Identify relationships between Item_MRP, Outlet_Type, and Item_Outlet_Sales.
• Baseline Models: Linear Regression, Decision Tree Regressor
• Advanced Models: Random Forest, XGBoost, LightGBM, CatBoost, Neural Networks (MLP)
• Evaluation Metric: Root Mean Squared Error (RMSE)

---

📂 Project 2: Disaster Classification

📖 Task Description
Given an image as input, the task is to classify the image into one of four disaster categories:

• CYCLONE
• EARTHQUAKE
• FLOOD
• WILDFIRE

📊 Dataset Information

• Training Samples: 400 per category
• Validation Samples: 100 per category
• Test Samples: 100 per category
• Dataset Link: Google Drive Dataset

Note: The dataset does not include .csv/.tsv/.txt annotation files. Images are organized in category-named subfolders; this folder structure can be used for labeling.

🤖 Recommended Approaches & Models

• Data Preparation: Image augmentation (rotation, flipping, scaling) and normalization.
• Baseline Models: Custom CNN architectures.
• Pretrained Models (Transfer Learning): ResNet50, VGG16/VGG19, EfficientNet, MobileNetV2.
• Evaluation Metrics: Precision, Recall, and F1-score per category; Macro Precision/Recall/F1 overall.

---

📂 Project 3: Handwriting Recognition

📖 Objective
Build and evaluate a model that recognizes handwritten text using labeled character images.

📊 Dataset Information

• Dataset: Kaggle – Handwriting Recognition
• Structure: train.csv, validation.csv, test.csv + image folders
• CSV columns:
  • filename: image filepath
  • identity: label (text/character)

⚙️ Workflow Overview

1. EDA: Visualize character distribution and sample images with labels.
2. Preprocessing: Normalize images, split into train/val/test, apply augmentation.
3. Model Selection:
   • Baselines: Logistic Regression, SVM (on flattened features)
   • Deep Learning: CNNs (e.g., simple CNN stacks/LeNet)
   • Advanced: CNN + RNN hybrids (CRNN) for sequence modeling
4. Training: Track accuracy/loss, use dropout, early stopping, regularization.
5. Optimization: Tune hyperparameters; compare architectures.
6. Evaluation: Character-wise F1 scores, confusion matrix, analyze misclassifications.

🤖 Recommended Approaches & Models

• Baseline: Logistic Regression, SVM
• Deep Learning: CNNs; CNN+RNN for sequences
• Modern: Transformer-based OCR approaches
• Metrics: Character-level F1, accuracy, confusion matrix

---

📂 Project 4: Robert Frost's Poem Generation

📖 Description
Generate new poems inspired by Robert Frost’s style using next-word prediction.

📊 Dataset Information

• Source: Project Gutenberg – Robert Frost’s Poems
• Preprocessing: tokenize words, build vocabulary, convert to sequences
• Split: 80% training / 20% testing

⚙️ Model Architecture

• Baseline: BiLSTM
  • Input: sequence length & vocab size
  • Hidden: BiLSTM units + dropout
  • Output: softmax for next-word prediction
• Training: categorical cross-entropy, Adam, batch size & epochs configurable

🤖 Recommended Approaches & Models

• Baseline: BiLSTM next-word model
• Advanced (Optional): GPT/T5 fine-tuning; experiment with diffusion-style text models

📈 Evaluation

• Perplexity (next-word prediction quality)
• Accuracy (coherence proxy)
• Generate example poems and qualitatively assess style/fluency

---

📂 Project 5: Cuisine Prediction

📖 Description
Predict the cuisine of a recipe from its list of ingredients (e.g., Indian, Mexican, Moroccan, Korean, Greek).

📊 Dataset Information

• Dataset: Google Drive – Recipe Dataset
• Format: JSON
• Fields:
  • id: unique recipe identifier
  • cuisine: target label (train only)
  • ingredients: list of ingredients

🔍 Example (train.json)

{
  "id": 24717,
  "cuisine": "indian",
  "ingredients": [
    "turmeric",
    "vegetable stock",
    "tomatoes",
    "garam masala",
    "naan",
    "red lentils",
    "red chili peppers",
    "onions",
    "spinach",
    "sweet potatoes"
  ]
}

🤖 Recommended Approaches & Models

• Baseline: ANN / simple feed-forward neural network with bag-of-words or TF-IDF features
• Intermediate: Embedding layers with CNN or BiLSTM for ingredient sequences
• Advanced (Optional): Transformer-based models

📈 Evaluation

• Accuracy (For simplicity)

---

✈️ Flight Delay Prediction

📖 Description

The goal of this project is to predict whether a flight will be delayed based on historical and contextual flight data.
This is a binary classification problem (Delayed vs On-time).

📊 Dataset Information

• Dataset Link: Google Drive Dataset
• Dataset: Flight records with the following key fields:
  • Year, Month, Day → flight date
  • DayOfWeek → numeric day of the week
  • Airline → airline carrier code
  • FlightNum → flight number
  • Origin → origin airport
  • Dest → destination airport
  • DepTime → actual departure time
  • ArrTime → actual arrival time
  • DepDelay → departure delay in minutes
  • ArrDelay → arrival delay in minutes (target variable)

🎯 Prediction Task

• Target Variable: ArrDelay
• Convert into binary classification:
  • Delayed → if ArrDelay > 15 minutes
  • On-time → otherwise

📈 Evaluation

• Metrics: Accuracy, Precision, Recall, F1-score
• ROC-AUC can also be used for better evaluation of class imbalance.

🤖 Recommended Approaches & Models

• Baseline: Logistic Regression / Random Forest
• Intermediate: Gradient Boosted Trees (XGBoost, LightGBM, CatBoost)
• Advanced:
  • Neural Networks with embedding layers for categorical features
  • Sequence models (RNNs/LSTMs) for temporal flight patterns