ARIMA vs LSTM Time-Series Forecasting

Simulation Setup

Stock Price CSV Data

Choose file or drag here

Expecting Date (DD/MM/YYYY) and Price columns

Look-back Window (Days)

Tuning Intensity

EXECUTION LOG IDLE

Ready to load dataset. Upload a CSV or load demo data to start.

Main Analysis Studio

Forecasting analysis completed successfully! Scroll down to inspect the model metrics.

ARIMA(1,1,1) Baseline

MSE: -

MAE: -

RMSE: -

LSTM Optimal Network

MSE: -

MAE: -

RMSE: -

Comparative Edge

Delta Improvement

PFE Forecasting Analysis Summary

This study compares the classical linear parametric model, **ARIMA(1,1,1)**, against the complex non-linear sequence modeling capabilities of an **LSTM Recurrent Neural Network**. The models were evaluated using 1-step-ahead rolling predictions (teacher-forcing style) on the final 10% test split.

The LSTM structure was optimized using a simulated **Hyperband tuner**, pruning suboptimal weights early on validation loss constraints. Classical ARIMA excels in low-volatility series but often suffers from lag when structural breaks occur, whereas LSTM learns temporal patterns and non-linear shifts.