Accurate and reliable price discovery mechanisms are crucial to all capital markets. Google Cloud has shared insights on how analytics can be derived from financial time series data, in real-time, using Machine Learning. In particular, a Keras model implementing an LSTM neural network for anomaly detection is provided. The financial data used comes directly from the Band Protocol public dataset available in Google BigQuery.

More interestingly, the data derived from the auto encoder-decoder could be considered a dataset in and of itself. This dataset can then be used to support the creation of a new decentralized oracle through custom oracle scripts on BandChain — further expanding the capabilities and offerings provided by Band Protocol.

On-chain smart contracts on any blockchains supported by Band oracles will then be able to in turn have access to pre-trained neural network and anomaly detection systems to perform complex business logic in a trustless manner without relying on other extra external parties.

Take a decentralized insurance protocol as an example, claims can be triggered and conditioned based on data anomaly detection while computation on-chain remains relatively minimal with little overhead. The idea can be further generalized to different types of software to enable smart contracts to delegate complex or expensive computations to GCP and thus creating hybrid cloud-blockchain applications.

Diving Into the Technicals

The whole process can be broken down into two parts:

* Part 1: Metrics calculation using Dataflow. The following metrics and technical indicators are included: Relative Strength Index (RSI), Moving Average (MA), and Open, High, Low, Close (OHLC).

* Part 2: Anomaly detection. We are using an LSTM model implemented in Keras.

Check out Google Cloud’s article for the full breakdown of technical details