For: DPG Media – Kiosk.nl marketing team
Project Goal
The primary goal of this project was to identify anomalies in kiosk performance data, specifically focusing on daily sessions and conversion metrics across different channels. By detecting unusual patterns, this analysis helps pinpoint potential issues or opportunities for improvement in the kiosk’s operation. The project also aimed to provide a clear visualization of these anomalies over time.
How it was built
This project began by loading and preprocessing time-series data, which included renaming columns and converting date fields. The core of the analysis involved several key steps: First, we engineered features such as conversion rate, rolling means, day of the week, and month. Then, we used a seasonal decomposition to capture trends and seasonality for each channel individually. An Isolation Forest model was employed for anomaly detection, trained separately for each channel to account for unique performance characteristics. The model was fit on a feature set including sessions, conversions, trends, and seasonality, and predictions were made to identify anomalous data points. Finally, the results were visualized using Plotly to display session activity and anomalies per channel over time.
Technologies used
- Python: Core programming language for data processing and analysis.
- Pandas: Used for efficient data manipulation and preprocessing.
- scikit-learn: Provides the Isolation Forest model for anomaly detection.
- statsmodels: Used for seasonal decomposition of time series data.
- Plotly: Used for interactive data visualization.