Time-Series Forecasting in Python: Predict Daily Restaurant Customers (Part 3)

Описание: Video 3 — Part 3: Random Forest, Linear Regression, LSTM & Feature Refinement for Peak Accuracy

Overview
We advance the forecasting pipeline in JupyterLab (online) by benchmarking three additional models—Random Forest, Linear Regression, and LSTM (Long Short-Term Memory)—against our prior XGBoost baseline. Using the top-performing features identified earlier, we refine model inputs to maximize predictive power and reduce forecast error. Each model is trained on the same time-aware split to ensure fairness and comparability. Performance is assessed rigorously using RMSE, MSE, and MAE, with direct comparisons and visual diagnostics to confirm gains in generalization and stability.

Watch the series
Paper Review (context first):    • Review Research Paper: Forecast Restaurant...  

Part 1 — Data Prep & Features:    • Time-Series Forecasting with XGBoost in Py...  

Part 2 — XGBoost & Evaluation:    • How We Beat the World's Best Forecasters U...  

This video — Part 3 (Model Comparisons & Feature Refinement):    / replace_with_part3  

What you’ll learn

How to train and tune Random Forest, Linear Regression, and LSTM for time-series forecasting

Proper use of best features (lags, rolling stats, autocorrelation, holidays, weather) to boost accuracy

Comparative evaluation using RMSE, MSE, and MAE across all models

How feature optimization impacts performance across linear, ensemble, and neural models

Visualization of model residuals and error trends to identify bias or underfitting

Final model ranking and interpretation of trade-offs between accuracy, complexity, and interpretability

Data & Features (context)
Daily customers (2019–2024), self-service restaurant
Lags (1/7/14/28), rolling stats, exponential smoothing, autocorrelation
Calendar & holiday effects (before/after), weather enrichment

Chapters
00:00 Recap & objective (why multiple models)
03:15 Random Forest: setup & training
07:20 Linear Regression: baseline comparison
10:45 LSTM: architecture & sequence design
14:30 Feature optimization & top-signal subset
17:00 RMSE/MSE/MAE comparison & visual analysis
20:00 Summary: accuracy gains, final takeaways

Tags
#randomforest #linearregression #lstm #timeseries #forecasting #python #machinelearning #demandforecasting #restaurant #pandas #tensorflow #scikitlearn #rmse #mae #mse #jupyterlab