csg
Résumé ↗
Menu
← All projects

Machine learning / Learning experiment

Stock Prediction Model

A study in learning from price history.

A TensorFlow notebook exploring price prediction, feature scaling and neural-network tuning on historical stock data.

TensorFlow · Keras Tuner · Python

View repository ↗

01 / THE PROBLEM

What needed solving.

Time-series prediction is a useful setting for learning preprocessing, tuning and evaluation. This notebook explores how a dense neural network maps historical features to a price target.

02 / FOLLOW THE WORKFLOW

See how it fits together.

A simplified example of the documented flow. Sample outputs are authored illustrations, not live model predictions.

Stock Prediction ModelIllustrative workflow

Tata Global Beverages dataset

Open · High · Low · Close

Start with historical OHLC values.

Documented 80 / 10 / 10 split

Training → validation → test

Scale features and partition the data.

Keras Tuner

Dense layers + Adam + early stopping

Search neural-network configurations.

Conceptual comparison, no forecast shown

Predicted price ↔ observed price

Inspect predictions against held-out values.

03 / ENGINEERING

The decisions behind it.

Keep tuning separate from testing

The README describes training, validation and test splits. Keras Tuner selects network width, depth and learning rate using validation behavior.

Use a compact modelling pipeline

MinMaxScaler prepares inputs. Dense ReLU layers, a linear output and Adam optimize mean squared error, with early stopping.

Inspect the comparison

Predicted-versus-actual visualizations help diagnose behavior. A stronger next evaluation would explicitly verify chronological splits, scaling scope and a naive forecasting baseline.

04 / EVIDENCE & LIMITS

Claims with context.

TensorFlowDense neural-network experiment
Keras TunerArchitecture and learning-rate search

The README documents the dataset, modelling setup and a predicted-versus-actual plot. It does not establish a reproducible held-out error or a comparison to a persistence baseline, so no accuracy or return claim is made.

Current limits & next validation

The notebook is a learning experiment, not a trading tool. Time ordering, leakage checks and a naive baseline are necessary before interpreting forecasting quality. The portfolio example contains no live financial data or investment recommendation.

05 / READ THE SOURCE

Go deeper.

Based on public repository documentation reviewed on 7 November 2026. Results are attributed to the project’s recorded evaluations; they were not rerun for this portfolio.

NEXT PROJECTTime Agent

TOOLS BEHIND THE WORK

Explore the stack. Drag the ribbon or select a tool.