Machine Learning and Government Data Analytics
Machine Learning and Government Data Analytics
In the world of government data analytics, machine learning can help government leaders get exponentially more value out of their data. From identifying fraud, waste and abuse to forecasting surges of new COVID variants, machine learning enables government agencies to deliver better outcomes for residents and business owners.
Like any business intelligence or data analytics initiative, the first step in developing machine learning solutions is clearly defining the business objectives. When working with our government customers on machine learning solutions, we dive in with a Discovery Session and start by asking these two questions: “What would you like your data to do?” and “What can your data do?” Based on the quality and characteristics of the data, as well as the desired outcome, we can determine the best machine learning approach.
Types of Machine Learning Solutions
There are a variety of machine learning approaches, but a government agency’s objective will inevitably be achieved by one (or some combination) of the following ML tasks:
Supervised Machine Learning
In supervised machine learning, the human developer acts as the trainer and teaches the system by feeding it training data. Essentially we show the system the “answer key” with inputs (questions) and outputs (correct answers) so it can learn the patterns and eventually infer the right answers on its own.
This approach is used when we want to better understand the relationship between two or more variables so that we can more accurately predict future outcomes. Use cases may include predicting whether a patient has a specific medical diagnosis, financial forecasts, weather forecasts, or population growth prediction.
Unsupervised Machine Learning Tasks
Unsupervised machine learning tasks are applied to a more challenging situation – when the human behind the system doesn’t know exactly what they are looking for. In this case, there are many possible relationships and associations within the data, and we need the machine learning solution to help us identify patterns.
Examples of unsupervised machine learning tasks include audience segmentation for targeting communications, anomaly/fraud detection, gene clustering, big data visualization, data compression, and recommendation systems.
Reinforcement Machine Learning
The objective of reinforcement machine learning is to develop a system that will, over time, learn to choose a set of actions that will maximize the likelihood of the desired outcome. The system learns by trial-and-error and will eventually be able to accurately predict which actions will lead to the optimal result.
Reinforcement machine learning can help organizations assess policies and programs – which addiction treatment methods are the most effective, which education programs result in higher student performance, which public safety initiatives actually decrease crime, etc.
Getting Started with Machine Learning
Government organizations looking to leverage the power of AI in their data analytics projects should adhere to these three best practices:
1. Ensure You Have High Quality, Reliable Data Sets
To take advantage of machine learning, governments must standardize and clean their raw data so that it can be ingested by the system. As part of this process, the data and analytics team will also need to ensure that individual privacy rights are protected.
This isn’t just about merging data, but rather performing quality checks to ensure it’s ready for use. Automation can play a crucial role here, as it removes human error from the data collection and organization process.
Data teams should also think about what data exists that they may not be taking advantage of. Because AI can bring together many disparate data sets, governments may be able to glean insights from unusual places. For example, in one project, researchers used Facebook data to understand the aftermath of natural disasters.
2. Emphasize Collaboration
Data is the fuel for an AI or ML model, and success is built over time. A pilot program should act as a feedback loop that allows the data team to evaluate initial results, add more datasets, and adjust the model as needed.
Such collaboration is best supported by a platform where everything from data ingestion to model building can happen in one place. Often government teams run into roadblocks when they try to hand off different parts of the project, thanks to disparate tools and processes. Government agencies should consider platforms that bring data prep, experimentation, and production into a seamless workflow, and ones that let users automatically track experiments and code. This will allow for faster deployment of models that can automate repetitive tasks for employees and model public health risks that inch you closer to the gold standard of predictive analytics.
3. Avoid the Black Box Problem
Governments must ensure any insights that are generated by ML models can be explained to a lay audience. Many people are reluctant to rely on AI and ML for decision making because they don’t understand what went into the model. This is sometimes called the “black box” problem.
One solution is to rely on open source software, which is built on transparency. It not only enables the aforementioned iteration and collaboration, but it makes it easier to explain how you came to particular results. This is particularly important in government, where agencies must not just make quick decisions, but maintain communication and trust with constituents.
In addition, as data scientists participate in the continuous feedback loop and watch the model learn, the results will become more reliable and the practitioners will be more comfortable with the entire AI/ML process.