https://facebook.github.io/prophet/docs/installation.html
In 2017, Facebook made its time-series forecasting tool, Prophet, open-source. This tool has the ability to produce accurate forecasts similar to those made by skilled analysts, but with minimal human effort. The Facebook Prophet is available in the form of an API in both Python and R.

Prophet works by using Additive Regressive models, which consist of four components:

1. Trend (g(t)): This component represents a piecewise linear or logistic growth curve trend. Prophet automatically detects changes in trends by selecting change points from the data.

2. Yearly Seasonality (s(t)): This component models the yearly seasonal pattern using the Fourier series. Additionally, it also considers the weekly seasonal component using a dummy variable.

3. Holidays (h(t)): This component allows users to provide a list of important holidays that may impact the forecast.

4. Error term (εt): This term represents the residual or error in the forecast made by Prophet.

The advantages of using Facebook Prophet include:

1. Accuracy: The Prophet tool is optimized for business-related problems encountered at Facebook and can generate forecasts as accurate as those made by skilled analysts. Moreover, it can produce results within seconds.

2. Data Processing: Facebook Prophet requires minimal data processing and can handle outliers and null values effectively.

3. Customization: Users have the flexibility to manually add seasonality and holiday values, allowing for easy integration of domain knowledge.

In this post, we will demonstrate the usage of Facebook Prophet with Python. Specifically, we will attempt to forecast the share price of Amazon Stock from 2019 to 2020 using historical share price data from 2015 to 2019.