QuestpondML.Lab1 🚀

A comprehensive Machine Learning laboratory project implementing various time series forecasting algorithms for financial data prediction, specifically targeting the Nifty 50 stock index.

📋 Table of Contents

• About
• Features
• Algorithms Implemented
• Project Structure
• Prerequisites
• Installation
• Usage
• Data
• Model Performance
• Contributing
• License
• Author

🎯 About

QuestpondML.Lab1 is an educational and experimental platform developed by Mrugendra (Yogesh) Bhure as part of the Questpond AI/ML Cohort. This project demonstrates practical implementation of multiple machine learning algorithms for financial time series forecasting, with a focus on real-world stock market prediction using the Nifty 50 index.

The project serves as a comprehensive learning resource for:

• Financial time series analysis and forecasting
• Multiple ML algorithm comparison and evaluation
• Feature engineering for time series data
• Model evaluation using comprehensive metrics
• AutoML experimentation and automated model selection

✨ Features

🤖 Machine Learning Algorithms

• Fast Forest Regression - Ensemble method for non-linear pattern detection
• Singular Spectrum Analysis (SSA) - Advanced time series decomposition with confidence intervals
• Support Vector Regression (SVR) - Kernel-based regression approach
• AutoML Experimentation - Automated model discovery and optimization

📊 Data Processing

• Lag Feature Engineering - Creates temporal features (6 lag periods)
• Relative Strength Index (RSI) - Technical indicator for momentum analysis
• Data Preprocessing - Custom parsing and normalization
• Train/Test Splitting - Proper temporal validation
• Logarithmic Transformation - Optional scaling for exponential patterns

📈 Evaluation Metrics

• R² (Coefficient of Determination)
• Mean Absolute Error (MAE)
• Mean Squared Error (MSE)
• Root Mean Squared Error (RMSE)
• Mean Absolute Percentage Error (MAPE)

🧠 Algorithms Implemented

1. Fast Forest Regression

• Type: Ensemble method using decision trees
• Best for: Non-linear relationships and complex patterns
• Features: 6 lag features as input, 95/5 train-test split

2. Singular Spectrum Analysis (SSA) ⭐ Currently Active

• Type: Time series decomposition method
• Best for: Advanced forecasting with confidence intervals
• Configuration: Window size: 12, Series length: 120
• Output: Point forecasts with upper/lower confidence bounds

3. Support Vector Regression (SVR)

• Type: Kernel-based regression
• Best for: High-dimensional feature spaces
• Features: 6 lag features + RSI, feature normalization, 6-period holdout validation

4. AutoML Experimentation

• Type: Automated model discovery
• Optimization: R² metric maximization
• Runtime: Limited to 120 seconds for efficiency

📁 Project Structure

QuestpondML.Lab1/
├── 📄 README.md                        # This file
├── 📄 LICENSE.txt                      # MIT License
├── 📄 QuestpondML.Lab1.sln             # Solution file
│
└── 📁 QuestpondML.Lab1/                # Main project
    ├── 📄 Program.cs                   # Main entry point
    ├── 📄 QuestpondML.Lab1.csproj      # Project configuration
    │
    ├── 📁 Labs/                        # ML implementations
    │   └── 📄 NiftyEstimator.cs        # Core ML algorithms
    │
    ├── 📁 Model/                       # Data models
    │   └── 📄 HistoricalStockPrice.cs  # Data structures
    │
    ├── 📁 Data/                        # Dataset
    │   └── 📄 Nifty 50 Historical Data.csv
    │
    ├── 📁 bin/                         # Build output
    ├── 📁 obj/                         # Build intermediates
    └── 📁 .vs/                         # VS configuration

🔧 Prerequisites

• .NET 8.0 SDK or higher
• Visual Studio 2022 or compatible IDE
• Windows OS (project uses Windows-specific paths)
• Git for cloning

📦 Installation

1. Clone the repository

   git clone https://github.com/anvaya/QuestpondML.Lab1.git
   cd QuestpondML.Lab1s

2. Restore NuGet packages

   dotnet restore

3. Build the project

   dotnet build

Package Requirements

Package	Version	Purpose
Microsoft.ML	5.0.0	Core ML.NET framework
Microsoft.ML.AutoML	0.23.0	Automated ML capabilities
Microsoft.ML.Mkl.Components	5.0.0	Intel Math Kernel Library optimizations

🚀 Usage

Running the Application

1. Navigate to the project directory

   cd QuestpondML.Lab1

2. Run the current experiment

   dotnet run

3. Switch between algorithms (edit Program.cs):

   // Currently active: SSA
   estimator.RunSsa();
   
   // Uncomment to use other algorithms:
   // estimator.RunAutoMLExperiment();
   // estimator.RunSVR();
   // estimator.RunReplWithFastForrest();

Running Specific Experiments

Fast Forest Regression

var estimator = new NiftyEstimator();
estimator.RunReplWithFastForrest();

Singular Spectrum Analysis

var estimator = new NiftyEstimator();
estimator.RunSsa();

Support Vector Regression

var estimator = new NiftyEstimator();
estimator.RunSVR();

AutoML Experiment

var estimator = new NiftyEstimator();
estimator.RunAutoMLExperiment();

📊 Data

Dataset Description

• Source: Nifty 50 Historical Data
• Format: CSV (Comma-separated values)
• Columns: Date, Price
• Date Format: dd-MM-yyyy (e.g., 01-05-2024)
• Time Range: May 2024 to November 2025 (monthly data)
• Price Format: Comma thousands separator (e.g., 22,901.35)

Data Schema

Date,Price
01-05-2024,22,901.35
01-06-2024,23,110.45
...

Enhanced Feature Set

The HistoricalStockPrice model includes the following features:

• Date: DateTime object for temporal ordering
• Price: Current price (log-transformed for modeling)
• LagPrice0-5: 6 lagged price features for temporal patterns
• RSI: Relative Strength Index (14-period default, configurable)

Preprocessing Pipeline

1. Date Parsing: Converts dd-MM-yyyy format to DateTime
2. Price Normalization: Removes comma separators, converts to float
3. Lag Feature Creation: Generates 6 lagged price features
4. RSI Calculation: Computes Relative Strength Index with configurable period (default: 6)
5. Optional Log Transformation: Applies logarithmic scaling for model training

🔧 Technical Implementation: RSI Calculator

The CalculateRSI function implements the standard Relative Strength Index algorithm:

Algorithm Details:

• Default Period: 6 periods (configurable, optimized for monthly data)
• Formula: RSI = 100 - (100 / (1 + RS)) where RS = Average Gain / Average Loss
• Smoothing: Uses smoothed moving average for RSI calculation
• Log Scale Handling: Automatically converts from logarithmic price data
• Edge Cases: Handles insufficient data gracefully (RSI = 0 for initial periods)

Features:

• Configurable calculation period
• Support for log-transformed price data
• Efficient memory usage with in-place calculations
• Robust handling of edge cases (zero losses, insufficient data)

📈 Model Performance

Evaluation Metrics Output Example

=== Model Performance Metrics ===
R² Score: 0.8567
MAE: 245.32
MSE: 85,432.11
RMSE: 292.27
MAPE: 1.24%

SSA Forecast Example

=== SSA Forecast Results ===
Period 1 Forecast: 23,450.67
Confidence Interval: [23,100.12, 23,801.22]

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request. For major changes, please open an issue first to discuss what you would like to change.

Development Guidelines

• Follow C# coding conventions
• Add XML documentation for new methods
• Update this README for significant changes
• Ensure all NuGet packages are updated

📝 License

This project is licensed under the MIT License - see the LICENSE.txt file for details.

👨‍💻 Author

Mrugendra (Yogesh) Bhure

• Role: Questpond AI/ML Cohort Member
• Focus: Machine Learning & Financial Time Series Analysis
• Project: QuestpondML Laboratory Experiments

---

🎓 Educational Purpose: This project is part of the Questpond AI/ML Cohort training program and serves as a practical demonstration of machine learning concepts applied to financial forecasting.

⚠️ Disclaimer: This project is for educational purposes only. The predictions and models should not be used for actual trading or investment decisions.