🤖 AI-Powered Forecasting (EXPERIMENTAL)

Predictive Analytics Dashboard

Advanced machine learning forecasting powered by Meta Prophet, XGBoost, Random Forest, Gradient Boosting, DBSCAN, and LSTM models with 78-feature enhanced intelligence

Experimental Feature: These predictions are generated by machine learning models and have not been independently verified by Belun. Results should be used for informational purposes only and supplemented with expert analysis.

Prediction Data is Outdated: ML forecasts were last updated 10.7 days ago The models below show historical methodology. For current predictions, please refresh the data from Google Colab.

Checking...

30-Day Forecast

Prophet ML Model

Checking...

Risk Classification

XGBoost Model

Checking...

Geographic Hotspots

DBSCAN Clustering

Checking...

Early Warnings

Anomaly Detection

30-Day Violence Forecast Model

Meta Prophet ML analyzes historical patterns, seasonal trends, and 7 key variables to predict daily incident rates for the next 30 days. Confidence intervals (95%) show the range where actual incidents are likely to fall.

Data Update Needed: This forecast was generated on 2025-10-04 14:33:23 (256.7 hours ago). The chart below shows historical predictions to demonstrate the model's methodology. Click "Refresh ML Data" to generate fresh predictions for the next 30 days.

How to Interpret: Rising trend lines indicate periods requiring increased security presence. Wide confidence intervals mean higher uncertainty. When updated regularly, this forecast helps plan resource allocation weeks in advance.

Risk Assessment

Geographic Hotspot Detection: Where Violence Clusters

DBSCAN clustering algorithm identifies municipalities with concentrated violence patterns. Hotspots = areas where incidents cluster together geographically and temporally, indicating systemic issues.

Strategic Priority: Hotspot zones require sustained intervention, not just incident response. Clustered violence indicates underlying tensions (land disputes, clan conflicts) that need community-level mediation and prevention programs.

Municipality Risk Scores

High Priority Zones

Risk Classification: Machine learning model (99.5% accurate) that categorizes each incident's severity level based on casualties, weapons used, response time, and 75 other factors. Helps prioritize which incidents need immediate attention vs. routine follow-up.

How Are Incidents Categorized by Risk?

Distribution of historical incidents across risk levels: Minimal, Low, Medium, High, Critical

What Factors Predict High-Risk Incidents?

Top 10 variables the AI uses to classify incident severity - higher importance = stronger predictor

Anomaly Detection: Spotting Unusual Violence Patterns

Two AI algorithms work together: Isolation Forest finds days with unusually high incidents, LSTM Neural Network detects when actual incidents differ significantly from predicted patterns.

Early Warning System: Anomalies signal potential coordinated violence, gang activity, or escalating conflicts. High-severity alerts (critical/high) require immediate investigation. Multiple anomalies in short timeframes suggest organized violence campaigns.

Detection Summary

Recent Anomalies

Enhanced ML Intelligence Models: Deeper Pattern Analysis

Advanced 78-feature predictive models analyzing weapons, responders, demographics, and conflict factors

What Are Enhanced ML Models?

While the core models above (Prophet, Risk Classification, Hotspots, Anomaly Detection) provide broad forecasting and pattern detection, these Enhanced Models dive deeper into specific tactical questions that security forces and policymakers need answered:

"Which weapons are most deadly?" → Weapon Lethality Analysis: Which Weapons Cause the Most Deaths?
"Where should we deploy specialized units?" → Geographic Risk Assessment: Where Are Different Weapons Most Dangerous?
"Which age groups need intervention programs?" → Age Demographics in Violent Incidents: Which Age Groups Are Most Involved?
"What response strategies work best?" → Response Effectiveness: Which Interventions Work Best?
"Which incidents might trigger more violence?" → Escalation Risk Predictor: Which Incidents Will Get Worse?
"How dangerous is this situation right now?" → Multi-Weapon Severity Analysis: How Many Weapons = How Much Danger?

These models analyze 78 different variables including specific weapon types (katapel, handgun, bladed, etc.), responder types (PNTL, F-FDTL, traditional leaders), age distributions, and conflict factors (alcohol, land disputes, domestic violence).

Weapon Lethality Analysis: Which Weapons Cause the Most Deaths?

This model analyzes historical incident data to predict the likelihood of fatalities based on weapon type. Lethality rate = percentage of incidents involving each weapon that resulted in deaths.

Data Quality Note: The lethality rates below may seem counterintuitive (e.g., body parts showing higher rates than firearms). This is likely due to: (1) Firearms being present but not used as the killing weapon, (2) Small sample sizes for some weapons creating statistical anomalies, (3) Multiple weapon incidents where guns are recorded but death results from beating. The model shows statistical associations in the data, not necessarily causal weapon effectiveness. Use with caution and expert interpretation.

Interpretation: Higher lethality rates may indicate more violent confrontations (e.g., beatings in rage) rather than weapon effectiveness. Small sample sizes (n<50) should be treated as preliminary. Larger sample weapons (n>500) provide more reliable patterns.

Geographic Risk Assessment: Where Are Different Weapons Most Dangerous?

Analyzes the combination of location and weapon count to identify high-risk scenarios. Risk score (0-100) combines historical severity, frequency, and casualties for each municipality-weapon combination.

Operational Value: Helps deploy resources geographically based on weapon-specific threats. Identifies which municipalities need specialized tactical units for particular weapon types.

Age Demographics in Violent Incidents: Which Age Groups Are Most Involved?

Analyzes perpetrator and victim age patterns to identify high-risk demographics. Shows how different age groups interact with weapon types and severity levels.

Prevention Value: Guides age-targeted intervention programs and youth engagement strategies. Helps design prevention campaigns specific to demographics most at risk of involvement in violence.

Response Effectiveness: Which Interventions Work Best?

Measures the success rate of different responder types in preventing deaths and serious injuries. Success rate = percentage of incidents where the response prevented fatalities or achieved peaceful resolution.

Strategic Value: Identifies which responder types are most effective for different incident types. Guides decisions on who to dispatch first and when to escalate or de-escalate response protocols.

Escalation Risk Predictor: Which Incidents Will Get Worse?

Predicts the likelihood that an incident will escalate into a larger conflict or trigger retaliatory violence. Escalation probability = likelihood (0-100%) that an incident leads to follow-on violence within 30 days.

Early Warning System: High escalation probability requires immediate de-escalation efforts. Key factors include multiple weapons, networked perpetrators, and property damage. Early intervention prevents cycles of retaliatory violence.

Multi-Weapon Severity Analysis: How Many Weapons = How Much Danger?

Analyzes how the number of weapons in an incident correlates with casualties and overall severity. Severity score = weighted combination of deaths, serious injuries, and minor injuries.

Threat Assessment: More weapons typically means coordinated violence and higher casualties. This model helps assess threat level during initial incident reports. Incidents with 3+ weapons require enhanced tactical response.