Member-only story
4 High-Impact Time Series Forecasting Project Ideas
And no, we’re not talking about predicting the stock market
The stock market captivates countless data science enthusiasts. Its daily ups and downs seem to offer endless opportunities to beat the market through smart predictive modeling. Accurately predicting stock prices consistently proves extremely difficult, if not impossible. The efficient market hypothesis suggests share prices already reflect all publicly available information, leaving no edge for predictive modeling efforts.
In many ways, stock market forecasting is the MNIST of time series — an overused beginner example that offers little real insight.
Rather than competing against Wall Street analysts, choose time series problems where your machine learning models can demonstrate clear value.
The following four ideas are interesting alternatives to stock price forecasting for data scientists looking to solve real-world problems. For each idea, we’ll explore one or two datasets to use.
Forecast Electricity Load
Electricity load forecasting refers to predicting future electric power demand based on historical data. Load forecasts estimate the amount of electricity customers will consume over a specified time horizon ranging from a day to a year. More accurate load forecasts allow utility companies to make better decisions about purchasing electricity and infrastructure investments to meet future demand.
Electrical grid operators closely monitor demand to ensure supply meets peaks and troughs. Unexpected load spikes can crash entire grids while under-forecasting demand leads to expensive last-minute power purchases. There is major value in fine-tuning electricity load forecasts to optimize power generation and cost efficiencies. Sharpening electricity load forecasts, even by small margins, generates considerable cost savings and grid reliability enhancements. Optimized demand forecasts improve decisions on where and when to purchase wholesale power.
