Time Series Sampling, Resampling, Prophet Modeling, LSTM Forecasting, and Clustering with Python

Time‑series data, such as stock prices, daily weather, or household electricity consumption, are common challenges for data scientists. The article uses a London household electricity dataset (recorded every 30 minutes from November 2011 to February 2014) as a running example.

First, the data is resampled using pandas to change its frequency, e.g., from half‑hourly to weekly or daily, by applying data.resample('W').sum() or similar calls. Various resampling options and aggregation methods are shown.

Next, the Facebook Prophet library (released in 2017) is introduced for time‑series forecasting. After renaming columns to ds (date) and y (value), a Prophet model is created, the changepoint_prior_scale is set to 0.15, and a future dataframe (e.g., two weeks ahead) is generated for prediction. The resulting forecast includes confidence intervals and visual trend plots.

The tutorial then moves to deep‑learning forecasting with an LSTM‑RNN. After scaling the data with MinMaxScaler , the series is split into training (80 %) and testing (20 %) sets, reshaped to [samples, timesteps, features] , and fed into an LSTM network. Training loss curves and prediction plots demonstrate good fit on both training and test data.

Finally, hierarchical clustering is applied to the same dataset. By adding columns for day‑of‑year and hour‑of‑day, the data is clustered using Ward’s method, and a dendrogram visualizes the merging process, illustrating how samples group together based on temporal consumption patterns.