Stock Price Prediction
Deep Learning / Time Series
A forecasting system built with recurrent architectures — RNN, LSTM, BiLSTM and GRU — trained and compared on historical stock data.
- Year
- 2024
- Role
- Solo build
- Source
- Public repository
Technologies
- Python
- TensorFlow
- LSTM
- GRU
- Time Series
- Jupyter
The problem
Sequence models are usually demonstrated one at a time, which tells you nothing about whether the architecture choice actually mattered for the series in front of you.
The solution
Train four recurrent architectures on the same data under the same conditions and compare them directly, so the comparison is about the model rather than the preprocessing.
Architecture
How it is put together.
Data preparation
Historical price data is cleaned and windowed into sequences.
Model training
RNN, LSTM, BiLSTM and GRU models are trained on identical inputs.
Comparison
Predictions are evaluated against held-out data across architectures.
Key features
What it actually does.
- Four recurrent architectures on identical data
- Consistent windowing and preprocessing
- Direct evaluation across models
Engineering challenges
The parts that were hard.
Financial series resist prediction
The honest result of this kind of work is usually about how much signal is not there — which is worth measuring rather than hiding.
Outcomes
What came out of it.
Qualitative outcomes only — no invented benchmarks or metrics.
A controlled comparison across four sequence architectures
A reusable windowing and evaluation setup
Next project
BloodLink — Blood Donation Platform
A platform connecting blood donors with recipients, tracked in real time.
Let’s build
Want something like this for your business?
Tell me the problem and I will come back with an architecture, a scope and a timeline.