Anomaly detection and forecasting on operational data

Objectives

• Prepare an operational time series (resampling, missing values, seasonality, exceptional events) and document the choices.
• Produce a forecast at a given horizon with a baseline model and then an advanced one, and measure the forecast error with the appropriate metric.
• Detect anomalies with at least two methods (statistical and machine learning) and calibrate the threshold to a false-alarm rate the business accepts.
• Build a logged alert board, with the severity level and the person notified.
• Define the tracking indicator with the business, and the human validation protocol that applies before any action triggered by an alert.

Audience

Engineers and data analysts in the energy and utilities, industry, logistics, banking and microfinance, and telecoms sectors; technically minded operations managers.

Prerequisites

Having completed DATA-201, or being at ease with Python, pandas and the basics of scikit-learn (training, validation, metrics). Checked by a placement test. One laptop per participant, 8 GB of memory minimum; a notebook environment with preinstalled libraries provided by Yolea; for remote delivery, a stable connection of at least 5 Mbit/s, microphone and camera. Training datasets provided (consumption, network load, transactions); the participant's own history is accepted if it is anonymised, covering at least twelve months.

Programme

• Time-series processing
• Forecasting and measuring error
• Anomaly detection and calibration
• Designing an alerting system
• Framing indicators with the business

Format

In-company or open enrolment. 14 hours, 2 consecutive days of 7 hours; remotely, 4 sessions of 3 hours 30 minutes over two weeks. On site, live remote or hybrid; in-company on your own histories, or open enrolment; 8 to 12 participants. 25% input, 75% notebook workshops; one running project per participant, from the raw series to the alert board; a systematic comparison of a simple baseline against an advanced model. The tracking indicator is set on the first morning: anomalies detected before an incident, and forecast error.

Assessment

Beforehand, a placement test. During the course, check quizzes and a practical challenge (detect the anomalies injected into an imposed series; the result is measured in detections and false alarms). At the end, a presentation of the running project's alert board, a same-day feedback survey and a named certificate of course completion. Three weeks later, a follow-up check-in.