Why Time Series Analysis Still Matters in the Age of Generative AI and LLMs
With the rise of generative AI and large language models (LLMs), it’s easy to assume that traditional data analysis techniques, like time series modeling, have taken a backseat. After all, LLMs generate human-like text, solve coding problems, and even analyze data — so why should we still focus on time series?
The truth is, time series analysis remains a critical component of AI-driven decision-making, especially in industries where forecasting, anomaly detection, and real-time insights are essential. While LLMs excel at processing and generating language-based data, they struggle with temporal dependencies, trends, and seasonality — which is where time series methods shine.
Ahead of our upcoming one-day virtual event, Time Series Mastery: Hands-On Workshops on March 13th, we’d like to review why Time Series is still a must-know skill in data science and AI.
Join us for a deep dive into the latest advancements in time series forecasting! This one-day virtual event explores AI-powered forecasting methods, emerging foundation models like TimeGPT, practical techniques for improving forecast accuracy, and optimization strategies using genetic algorithms. Whether you’re looking to refine your forecasting approach or explore cutting-edge innovations, these sessions will provide the insights and tools needed to tackle real-world time series challenges with confidence.
1. Forecasting the Future — Something LLMs Can’t Do Well
Generative AI and LLMs are great at pattern recognition and information synthesis, but they’re not designed to predict future values based on past observations. In contrast, traditional time series models like ARIMA, Prophet, and LSTMs are built specifically for forecasting, making them indispensable for financial markets, supply chain planning, and demand forecasting.
2. Real-Time Decision-Making Needs More Than Just Language Models
Many industries — like energy, healthcare, and cybersecurity — rely on real-time data streams to make split-second decisions. Detecting anomalies in IoT sensor data, monitoring stock market trends, or predicting disease outbreaks requires specialized time series techniques. While LLMs can analyze reports after the fact, they lack the precision needed for real-time anomaly detection.
3. Temporal Dependencies Matter
Time series data isn’t just a collection of numbers — it’s sequential, time-dependent, and often highly structured. LLMs, despite their massive capabilities, don’t inherently understand time-based relationships the way traditional time series models do. This makes methods like exponential smoothing, Kalman filters, and RNNs essential for tracking long-term trends and seasonality in data.
4. The Best AI Systems Combine Both
Rather than replacing time series methods, LLMs and generative AI are complementary tools. For example, LLMs can help interpret and communicate insights from time series analysis, automate reporting, and even enhance feature engineering. But when it comes to accurate forecasting, detecting trends, and analyzing historical patterns, time series models remain the gold standard.
Join Us at the Time Series Analysis Mastery: Hands-On Workshops Event!
The “Time Series Mastery: Hands-On Workshops” is a one-day virtual event dedicated to exploring the fascinating world of time series analysis. The event features four insightful sessions led by industry experts, each designed to enhance your understanding and application of time series forecasting.
Optimizing Forecast Stability and Accuracy
Instructors: Jeff Tackes, Global Head of Forecasting at Kraft Heinz | Hamed Alikhani, Data Scientist at Kraft Heinz
The presenters will introduce a novel approach leveraging genetic algorithms to optimize both forecast stability and accuracy. The session addresses the common pitfall of month-over-month volatility in predictions, known as the whiplash effect, and demonstrates how this method systematically adjusts model weights based on historical deviations and performance metrics to solve key business challenges.
State of Foundation Models for Time Series Forecasting
Instructor: Marco Peixeiro, Applied AI Scientist at Nixtla
Marco Peixeiro covers the emerging field of foundation models in time series forecasting. The talk explores core concepts such as pretraining, transfer learning, and fine-tuning, and examines major contributions like TimeGPT, Chronos, Moirai, and TimesFM. A hands-on example with TimeGPT demonstrates how a foundation model can be used and how it compares to traditional methods.
Forecasting the Future Using Time Series
Instructor: John Mount, PhD, Principal Consultant at Win Vector LLC
John Mount shares a simplified problem notation that helps participants survey available solution offerings and succeed with time series packages in R and Python. The session emphasizes the importance of attributing and inferring hidden state in time series problems and deploying reliable and effective forecasts using standard open-source packages.
Unlocking the Future with AI-Driven Time Series Forecasting
Jeffrey Yau, Former Global Head of Data Science and Engineering at Amazon Music
In this engaging workshop, Jeffrey Yau explores how artificial intelligence has transformed time series forecasting. Participants will delve into classical forecasting models alongside modern AI-driven approaches, gaining both theoretical knowledge and hands-on experience. The session aims to demystify AI-driven forecasting, empowering attendees to apply these tools for actionable insights in their unique domains.
Sign me up!
These sessions collectively offer a comprehensive exploration of time series forecasting, from foundational models to advanced AI-driven techniques, providing attendees with practical skills and insights applicable across various industries.
Date: March 13th, 12:00 PM ET
Location: Virtual
Time may move forward, but time series analysis remains just as important as ever. Don’t miss your chance to stay ahead of the curve!
