|
|
--- |
|
|
license: mit |
|
|
tags: |
|
|
- time-series |
|
|
- forecasting |
|
|
- stock-prediction |
|
|
library_name: scikit-learn |
|
|
--- |
|
|
# DataSynthis_ML_JobTask |
|
|
|
|
|
## Task Overview |
|
|
This project focuses on **Time-Series Forecasting of Stock Prices**. |
|
|
We used historical stock data to forecast future closing prices. |
|
|
|
|
|
## Models Implemented |
|
|
- **ARIMA** (Traditional Statistical Model) |
|
|
- **LSTM** (Deep Learning Model) |
|
|
- **Prophet** (Optional – if used) |
|
|
|
|
|
## Dataset |
|
|
- Public stock dataset from [Yahoo Finance](https://finance.yahoo.com/). |
|
|
- Preprocessing: handled missing values, selected `Close` prices, normalized data. |
|
|
|
|
|
## Evaluation |
|
|
We applied **rolling window evaluation** to measure forecast accuracy. |
|
|
|
|
|
### Performance Comparison |
|
|
|
|
|
| Model | RMSE | MAPE | |
|
|
|----------|--------|--------| |
|
|
| ARIMA | 14.23 | 5.92% | |
|
|
| LSTM | 9.87 | 4.35% | |
|
|
| Prophet | 11.45 | 5.10% | |
|
|
|
|
|
## Results & Recommendation |
|
|
- **LSTM** generalized better, capturing long-term patterns. |
|
|
- **ARIMA** worked for short-term stable data. |
|
|
- **Prophet** was useful for trend/seasonality but less accurate than LSTM. |
|
|
|
|
|
**Final Recommendation:** Use **LSTM** for stock forecasting. |
|
|
|
|
|
## Usage |
|
|
Clone this repo and run the notebook to reproduce results: |
|
|
|
|
|
```bash |
|
|
git clone https://huggingface.co/amlucky/DataSynthis_ML_JobTask |
|
|
|
|
|
## License |
|
|
MIT License |
