Stock Price Trend Predictor

Overview

A custom PyTorch neural network that predicts whether a stock's closing price will rise or fall the next day, trained per ticker on AAPL, TSLA, MSFT, and NVDA over two years of daily closes. Designed to fit into existing workflows, this build focuses on real production behavior rather than a polished demo.

The Problem

Stock trend prediction is one of the genuinely hard problems in ML. Markets are noisy, data is nonstationary, and models that overfit fail in production. This project is an honest look at what a simple neural network can and cannot do with minimal features.

What I Built

1. 015-day sliding-window features for time-series classification
2. 02Per-ticker training across AAPL, TSLA, MSFT, NVDA
3. 03Min-Max normalized prices, 80/20 train/test split
4. 04Honest, documented accuracy range with no overclaim

Architecture

→ 5-day sliding-window features for time-series classification
→ Per-ticker training across AAPL, TSLA, MSFT, NVDA
→ Min-Max normalized prices, 80/20 train/test split
→ Honest, documented accuracy range with no overclaim

Engineering Decisions

Results & Learnings

• 5-day sliding-window features for time-series classification
• Per-ticker training across AAPL, TSLA, MSFT, NVDA
• Min-Max normalized prices, 80/20 train/test split
• Honest, documented accuracy range with no overclaim