lesleypotters / Forecasting-in-R

Forecasting

This repository contains a few forecasting projects written in R. Besides the general visualization and decompositions, several forecasting techniques have been applied:

• statistical methods, such as exponential smoothing, Holt, linear regression and theta
• Machine Learning methods, such as Deep Learning (Neural Network) and decision trees (XGBoost)

Sales data

• Data visualization
• Filling missing values
• Identifying outliers + normalization
• Visualize weekly and monthly series
• Decomposition of trend and seasonal components

Daily sales of a product sold in a store

• Data visualization
• Some measures for forecastability are computed: ADI and CV^2
• Empirical distribution of the demand of the product and compute some percentiles of number of daily sales
• Decomposition of seasonal, trend and random components

International airline passengers

• Decomposition in seasonal and trends components
• Generate forecasts using Simple, Holt and Damped Exponential Smoothing
• Measuring accuracy including prediction intervals

Power generation data (wind turbine)

• Plotting some data for better understanding
• Forecasting of power generation by applying a Linear Regression Model
• Forecasting of power generation by applying a Neural Network

Electricity prices

• Simple tests for missing values and outlier detection
• Normalizing and imputing these
• Correlation matrix
• Decomposition
• Statistical forecasting: ** insample/outsample data separation ** applying theta forecasting
• ML forecasting: ** scaling of data ** factorizing ** including data lag (1 week, 2 weeks) ** insample/outsample ** define NN model and hyperparameters ** define XGBoost model and hyperparameters
• Compare scores of all models