Case Study · Reproducible Trend Workflow

Meditation Trend Pulse

A reproducible workflow analyzing normalized Google Trends interest in meditation, mindfulness, and breathwork, with configurable updates, Prophet forecasting, and a multi-page Streamlit dashboard.

Streamlit Prophet Python PyTrends Pandas Automation Time Series Trend Analysis Google Trends

⌥ GitHub Repo

3Dashboard Pages

9Cleaned Datasets

ConfigurableUpdate Workflow

ProphetForecasting Engine

Project Access

Meditation Trend Pulse — Source and reproducibility notes

↗ Open GitHub Repo

The source repository documents the dataset snapshot, configurable workflow, dashboard code, and forecasting demonstration.

The Problem

Google Trends provides sampled, normalized search-interest scores rather than raw search volumes. The analytical challenge is to preserve that context while comparing topics over time, across countries, and through related-query signals.

The goal was to build a reproducible, configurable update workflow and dashboard for exploring those relative-interest patterns. The published dataset snapshot runs through September 28, 2025.

Approach

Configurable Data Updates

Built a Python updater using PyTrends to rebuild the project datasets when a new weekly observation is available. A local cron wrapper is included, but it requires environment-specific configuration and is not presented as an active scheduler.

PyTrendsAutomationPython ScriptsScheduling

EDA & Time Series Analysis

Cleaned and transformed 9 datasets covering global trends, country-level interest, and related queries. Applied smoothing, peak detection, percent change calculations, and monthly heatmaps to surface meaningful patterns in noisy trend data.

PandasNumPySmoothingPeak DetectionHeatmaps

Prophet Forecasting

Used Prophet to demonstrate 12-week and 52-week forecasts from the published snapshot, with uncertainty intervals and a 26-week holdout backtest for basic error evaluation.

ProphetTime SeriesForecastingUncertainty Intervals

Multi-Page Streamlit Dashboard

Built a 3-page Streamlit app with custom chakra-themed UI components, Altair charts, and a polished layout. Each page answers a distinct analytical question — global trends, country breakdowns, and related query analysis.

StreamlitAltairCustom UIMulti-page

Dashboard Pages

🌍

Global Trends

5-year search interest over time, smoothed trend lines, % change table, peak interest dates, and Prophet forecasts for each keyword.

🗺️

Country Trends

Normalized relative-interest scores by country and keyword, with cross-keyword comparisons that should not be interpreted as population demand.

🔍

Related Queries

Top and rising search queries associated with each keyword — surfacing what people are actually searching alongside meditation and mindfulness, including shared interest signals across keywords.

Key Findings

The dashboard compares normalized relative-interest indices for meditation, mindfulness, and breathwork over the published five-year window; the scores are not raw search counts.

Country views show where each topic receives a higher relative Google Trends score within the selected query set, geography, and time window—not per-capita interest or market demand.

Endpoint changes, peaks, and related-query growth are descriptive and can be sensitive to sampling, low starting values, and the selected dates.

Prophet outputs demonstrate possible 12-week and 52-week trajectories with widening uncertainty intervals; they are model estimates rather than guaranteed future growth.

Analytical Value

The project demonstrates ownership of ingestion, cleaning, analysis, forecasting, and dashboard delivery within a reproducible portfolio workflow.

It can support exploratory trend monitoring and help frame follow-up research questions, but Google Trends alone is not evidence of audience size, causal motivation, or commercial demand.

Interpretation & Limitations

Google Trends supplies a sampled, normalized 0–100 relative-interest index. Scores depend on the query set, geography, and time window; country values should not be added or interpreted as population-level demand.

Peaks, related queries, and regional patterns are descriptive associations and do not establish why people searched or prove causation. PyTrends is an unofficial interface, so future reruns may return slightly different sampled values.