RealityStream

RealityStream Forecasting Models

Our Machine Learning (ML) models join feature and target (Y=1) data based on location IDs.
Scope includes: Countries, States, County FIPS, Zip Codes, and Brain Voxels (for eye blinks).

• Run Models CoLab
• Random Forests for Healthy Bees
• Random Bits Forest for Eye Blinks

The Run Modles colab merges feature and target datasets in Pandas.
Training files can be reviewed by setting save_training=True.

Select one or more models, then paste the resulting yaml in the Run Models CoLab.

Logistic Regression (lr)

• Type: Linear model for binary classification (extendable to multiclass).
• Key Feature: Predicts probabilities using the logistic (sigmoid) function.
• Best for: Clean, balanced datasets with approximately linear relationships between features and target.
• Common Use: Medical diagnostics, marketing (e.g., churn prediction), financial risk.
• Limitations: Struggles with non-linear relationships and complex patterns.

Random Forest Classifier (rfc)

• Type: Ensemble of decision trees using bootstrapping and random feature selection.
• Key Feature: Reduces overfitting and variance through randomness and averaging.
• Best for: Diverse data types, handles outliers, missing data, and categorical features.
• Common Use: Fraud detection, bioinformatics, credit scoring.
• Limitations: Less interpretable, computationally expensive with large datasets.

Random Bits Forest (rbf)

• Type: Variation of Random Forests using bit-based transformations for feature splitting.
• Key Feature: Efficient handling of high-dimensional, binary, or sparse data.
• Best for: Cybersecurity, large-scale categorical or binary feature data.
• Common Use: High-dimensional binary datasets, one-hot encoded data.
• Limitations: May not perform as well on general-purpose or continuous feature datasets.

Support Vector Machines (svm) - Runs Slow

Runs slow (over an hour), best to avoid for now.

• Type: Supervised learning algorithm for classification and regression.
• Key Feature: Maximizes the margin between classes, supports non-linear relationships via kernel tricks.
• Best for: Small to medium-sized, high-dimensional datasets with well-separated classes.
• Common Use: Text classification, image recognition, bioinformatics.
• Limitations: Struggles with large datasets and noisy data, can be computationally expensive.

Neural Network Multi-Layer Perceptron (mlp)

• Type: Feedforward artificial neural network with multiple layers.
• Key Feature: Learns complex non-linear relationships through backpropagation.
• Best for: Large datasets with intricate feature interactions (e.g., images, speech).
• Common Use: Deep learning tasks, image processing, speech recognition.
• Limitations: Requires large datasets and computational resources, prone to overfitting on small datasets.

XGBoost (xgboost)

• Type: Gradient-boosted decision tree algorithm.
• Key Feature: Sequentially corrects previous errors, highly efficient with built-in regularization.
• Best for: Large, structured tabular data, handles missing data natively.
• Common Use: Financial modeling, fraud detection, machine learning competitions.
• Limitations: Complex and harder to interpret, requires tuning for optimal performance.

Continue

Paste the resulting yaml parameters list into the second step of the Run Models CoLab.