Predicting earthquakes, solar flares, landslides, and more.
Besides the obvious projects we hear so much about — segmentation, churn, recommendation engines, and so on — there are countless creative, underexplored opportunities to use predictive analytics.
Here are 20 exciting use-cases.
Immunotherapy is a promising way to shrink advanced tumors and prolong survival, but it can only be used for 20–30% of patients. It can have severe side effects, so we want to know ahead of time if patients would benefit.
Researchers at the Princess Margaret Cancer Centre created a study evaluating patients’ responses to immunotherapy via a test based on their tumor profile.
Response to immunotherapy can be predicted based on increasing or decreasing levels of DNA fragments shed from the tumor into the blood (called ctDNA). A decrease in ctDNA at 6–7 weeks after treatment with a specific treatment was associated with a positive response, while a rise in ctDNA levels was linked to rapid disease progression.
This is a great example of predictive analytics because there are tens of thousands of genes in each cancer, so there’s a huge range of potential mutations in different individuals.
Around 10% of women develop gestational diabetes during pregnancy. 30–50% of these women get type 2 diabetes (T2D) later in life. Scientists identified a metabolic signature that can predict, with over 85 percent accuracy, if a woman will develop T2D.
Patient non-compliance is a huge concern in the healthcare industry. When patients don’t participate in their medical care, outcomes suffer.
You can use data like how often patients use the Internet to predict medical care participation.
If more people had a simple test to predict their risk of getting a heart attack, lives would be saved. Coronary Calcium Scans show how much Calcium is in a person’s arteries, which can be used to predict the risk of a heart attack.
Half of people will get cancer. Researchers found that cancer treatment outcomes can be modulated by the levels of specific gut bacteria.
In other words, data on gut bacteria at the onset of cancer treatment can be used to predict how well people respond to anticancer drugs.
6. Skin cancer
Researchers at Seoul National University used 220,000 images of people with 174 skin diseases to build a neural network that classifies a skin disorder.
The results were impressive, and on-par with the accuracy of dermatology residents. In conjunction with the expert opinion of dermatologists, the predictive model can increase accuracy, acting as Augmented Intelligence.
Researchers analyzed artificial fault zones to identify processes that lead to earthquakes. Predictive modeling is used to predict the frictional strength of phyllosilicates, the movement of which is responsible for earthquakes.
Blood biomarkers found in severe COVID-19 patients could be used to predict the severity of a new COVID-19 case, potentially enabling more efficient resource allocation in hospitals.
Proteins are found in all our cells, playing an important role in blood coagulation, as the main constituents of hair and muscle, and more. Their function can be determined by their molecular structure. AI can be used to predict contact pairs, and thus the function of a protein, using language translation techniques that identify which amino acids form a pair.
Early Alzheimer’s diagnosis leads to better treatment. Researchers found a group of plasma signaling proteins that reflect changes in the brain during early Alzheimer’s.
Predictive analytics can be used on 18 proteins in blood plasma to identify patients with Alzheimer’s or to predict the onset of Alzheimer’s, with close to 90% accuracy.
AI can be used to analyze training and racing data by wearable fitness trackers to calculate a critical speed value that’s predictive of a runner’s marathon time with high accuracy.
Other data like the shoes you wear, whether you run with a pacer, the flatness of the course, and weather conditions also impact running time.
12. Suicide risk
Predicting individualized suicide risk could help reduce suicides — one of the leading causes of death — with targeted treatment.
Data from longitudinal electronic health record data from 3.7 million patients were used to predict suicide with roughly 70% accuracy.
13. Solar flares
Differences in gamma radiation can be used to predict solar flares more than a day in advance, giving advance warning to protect satellites, power grids, and astronauts from harmful radiation.
People are sharing millions of photos every day, including a treasure-trove of flower images. We can identify insects visible in these photos and compare these observations to visitation rates observed in controlled trials of the same plans to predict pollinator activity.
Predicting recessions is no easy task. This is clearly demonstrated by the fact that you’ll find thousands of predictions of an impending recession, every year.
One man, however, successfully predicted the political stress and recession of 2020, over 10 years ago. Check out how:How One Man Predicted 2020’s InsanityAn incredible case study on predictive analytics.medium.com
Some alien species have the potential for mass destruction by wiping out native plants and permanently altering ecosystems. However, only a small percentage of non-native insects damage out forests — so it’d be useful to predict whether a new insect is harmless or destructive.
We can use data on insect traits, host traits, evolutionary history, and more to fairly reliably predict destructiveness.
We’re creatures of habit. Historical location data from our smartphones, communication patterns, app usage, and other metrics can be used to predict our mobility patterns to an average of 20 meters error over 24 hours.
18. Tomorrow’s mood
Our mood has an important impact on our clinical health. Historical data on our mood, stress, and physical health, such as from wearables, can be used to predict our mood tomorrow.
Landslides can destroy whole communities in an instant, and they’re becoming a bigger and bigger problem due to climate change.
We can predict areas where landslides will occur to proactively prepare, using data like precipitation and underlying soil saturation.
20. Superbowl winner
Using data from FiveThirtyEight and Elo ratings of teams and quarterbacks, you can train a model to predict the margin of victory of a specific team, and the error of the model.
In a simulation of 10,000 games, researchers (correctly) predicted the Chiefs would win in 2020.Towards AI — Multidisciplinary Science Journal