PHAROS AI Factory project announced the 14th Course of its Training Series, with the title “Machine & Deep Learning for Time Series forecasting“, that successfully took place on September 15th, 2026.

Presentation language: Greek

Course Description: This hands-on session worked through a complete forecasting pipeline in Python related to time-series data. Participants began with preprocessing (handling missing values, outliers, stationarity, and feature engineering), then moved through classical machine learning models, gradient-boosted trees, and deep learning architectures including LSTMs, temporal convolutional networks, and Transformer-based models such as PatchTST. The course was built around realistic use cases, working with live coding alongside participants.

Audience: ML Engineers, Data Scientists, Academic Researchers, Business Owners

Learning Objectives

By the end of the course, participants were be able to:

• Prepare time series data for supervised learning (handling irregular sampling, missing values and outliers — while avoiding the data leakage patterns that invalidate most forecasting results).
• Select appropriate baselines and evaluation metrics, and apply walk-forward validation rather than standard cross-validation.
• Build and compare classical, gradient-boosting and deep learning forecasting models, and judge when the added complexity of deep learning is justified.
• Recognise the strengths and limitations of current architectures, from LSTMs and TCNs to Transformer-based and foundation models.

The course’s presentation material can be found here.

The course’s recordings can be found here.