📈 Financial Market Prediction Using Regression on Sequential Time-Series Data

A machine learning mini project that applies regression algorithms to real-world financial time-series data to predict continuous market price values.

---

📌 Overview

Predicting financial market movements is one of the most challenging problems in data science due to the noisy, non-stationary, and sequential nature of financial data. This project explores how classical regression algorithms can be applied to structured time-series financial data to forecast price-related values.

The pipeline covers everything from data loading and feature engineering to model training, evaluation, and visualization — giving a clear picture of where regression models succeed and where they fall short in real-world financial markets.

Built as a Mini Project during a 3-month Machine Learning Internship at VCodEZ (Nov 2025 – Feb 2026)

---

🚀 Key Features

• Time-Series Regression — Applied 4 regression models to sequential financial data
• Lag Feature Engineering — Created lag-based and rolling-window features to capture temporal patterns
• Temporal Train-Test Split — Avoided data leakage with time-aware splitting (no random shuffling)
• Multi-Model Comparison — Evaluated and compared Linear, Ridge/Lasso, Decision Tree, and Random Forest regressors
• Visualization — Matplotlib plots for actual vs predicted values, residuals, and feature importances
• Honest Evaluation — Clearly documents both capabilities and limitations of regression in financial markets

---

🛠️ Tech Stack

Technology	Version	Purpose
Python	3.9	Core language
Pandas	2.x	Data loading, manipulation, feature engineering
NumPy	1.x	Numerical operations
Scikit-learn	1.x	Regression models, metrics, preprocessing
Matplotlib	3.x	Plotting actual vs predicted, residuals
Jupyter Notebook	—	Development and experimentation environment

---

🧠 Regression Models Used

Model	Strength	Limitation
Linear Regression	Fast, interpretable, low variance	Fails on non-linear market patterns
Ridge / Lasso Regression	Handles multicollinearity, regularized	Still linear — limited expressiveness
Decision Tree Regressor	Captures non-linearity	Overfits easily on noisy financial data
Random Forest Regressor	Handles non-linearity + variance	Risk of overfitting on financial noise

---

📊 Evaluation Metrics

Metric	Description
MAE (Mean Absolute Error)	Average magnitude of prediction error
RMSE (Root Mean Squared Error)	Penalizes large errors more heavily
R² Score	Proportion of variance explained by the model

---

⚙️ Project Pipeline

Raw Financial Data (CSV)
        |
        v
1. Data Loading & Exploration (Pandas)
        |
        v
2. Feature Engineering
   - Lag features (t-1, t-2, t-3 ...)
   - Rolling mean / rolling std windows
   - Technical indicators (optional)
        |
        v
3. Time-Aware Train-Test Split
   (No random shuffling — preserves temporal order)
        |
        v
4. Model Training
   (Linear, Ridge, Lasso, Decision Tree, Random Forest)
        |
        v
5. Evaluation (MAE, RMSE, R²)
        |
        v
6. Visualization (Actual vs Predicted, Residuals, Feature Importance)

---

📁 Project Structure

Machine-learning/
├── Financial Market/
│   ├── read.md               # Project notes and model comparison summary
│   └── [notebooks / scripts] # Data processing, training, evaluation
├── README.md                 # Full project documentation
└── .gitignore

---

💡 Key Learnings

• Temporal dependency matters — Random train-test split leaks future data into training, leading to falsely high scores.
• Linear models are insufficient — Financial data is inherently non-linear; regularized linear models still underfit.
• Overfitting is real — Tree-based models capture training patterns well but generalize poorly on unseen market data.
• Feature engineering is critical — Lag features and rolling statistics are more informative than raw price values alone.
• Regression ≠ trading strategy — Predicting price direction and predicting exact price values are very different problems.

---

🔮 Future Scope

• Integrate LSTM / GRU deep learning models for better sequence modelling
• Add technical indicators (RSI, MACD, Bollinger Bands) as engineered features
• Build a multi-step forecast pipeline (predict t+1, t+2, ... t+n)
• Deploy as a Streamlit web dashboard for live market monitoring
• Experiment with XGBoost and LightGBM for gradient boosting on time-series

---

👨‍💻 Developer

Field	Details
Name	Mulamreddy Venkata Vasu Deva Reddy
GitHub	@MVVasudevreddy
LinkedIn	venkata-vasu-deva-reddy-mulamreddy-6666vdr
Institution	Bharath Institute of Higher Education and Research (BIHER), Chennai
Internship	Machine Learning Intern — VCodEZ, Chennai (Nov 2025 – Feb 2026)
Project Type	Mini Project — B.Tech CSE (2022–2026)

---

⭐ If this project helped you understand regression on time-series data, give it a star!

Department of Computer Science and Engineering | BIHER, Chennai