Santosh Palaskar, a 4th year Ph.D. student at IEOR department, IIT Bombay, who works with Prof. Narayan Rangaraj and Prof. Nandyala Hemachandra, presented his paper at the IAAI/AAAI-2024 conference in Vancouver, Canada during February 20th to 27th, 2024.
Paper title: AutoMixer for Improved Multivariate Time-Series Forecasting on Business and IT Observability Data
Abstract: Efficient business processes rely on key performance indicators (Biz-KPIs), which can suffer from IT failures. Business and IT Observability (BizITObs) data combines Biz-KPIs and IT events as multivariate time series. Forecasting Biz-KPIs in advance can boost efficiency and revenue by enabling proactive measures. However, noise in the data can hinder existing forecasting models. We introduce AutoMixer, which improves forecasting accuracy by 11-15% through channel-compressed pretraining with an AutoEncoder and integration with TSMixer. This fusion enhances TSMixer's potency and generalizes well across tasks, providing actionable insights.
