Time-series forecasting with LSTM — Deep Learning & Neural Networks

Time-series forecasting predicts future values based on historical sequences — predicting IPL run rates over the next 5 overs from the last 10, forecasting sensor readings, or estimating stock prices. LSTM is particularly well-suited for time-series forecasting because the cell state can maintain long-horizon context (what happened 20 overs ago may influence the current run rate pattern) while the hidden state captures recent trends. The two critical engineering decisions for time-series forecasting are: (1) sequence construction — how to transform a 1D time series into (samples, timesteps, features) format via sliding window, and (2) target formulation — single-step ahead prediction, multi-step direct prediction, or multi-step recursive prediction. Poor sequence construction is the most common source of data leakage in time-series forecasting: normalising the series before splitting train/test leaks future statistics into the training set. Understanding the correct preprocessing, model architecture, and evaluation protocol for time-series is essential for producing reliable forecasts that hold up in production.

Analogy🏏Cricket

🏏 Think of it like cricket: The Duckworth-Lewis-Stern (DLS) method, the ICC bowling economy metric, and the net run rate each measure team performance but optimise for completely different things — DLS cares about wickets-in-hand, economy cares about runs-per-over, NRR cares about tournament-wide margins. Using the wrong metric to evaluate a bowler is like using MSE for classification: technically computable but optimising the wrong thing entirely. Just as you would never rank a spinner by batting average, you should never use MSE when your model predicts probabilities. Just as focal DLS adjustments down-weight easy chases and amplify close finishes, focal loss down-weights easy examples and amplifies hard minority-class examples. The choice of scoring metric defines what excellence means — and so does the choice of loss function.