CRISIS PREVENTION

CRISP-DM v1.0

MODEL DEPLOYED
RECALL: 92%
PRECISION: 93%
NET VALUE: $721,535
CRISP-DM PIPELINE / BUSINESS UNDERSTANDING

BUSINESS UNDERSTANDING

CRISP-DM Phase 1 -- Problem definition, stakeholders, and success criteria

THE PAIN POINT
CLIENT
Health Insurance Providers
PROBLEM
Crisis therapy is expensive to cover compared to preventative care. Teen phone addiction is driving mental health crises that insurers pay for.
COST COMPARISON
$2,673/ crisis episode
vs. low-cost preventative care (counseling apps, wellness coaching)
DECISION SUPPORTED
When to trigger parent interventions once a teen crosses the tipping point
THE BUSINESS HOOK
5.32B
People worldwide using a cell phone
7.7B
Worldwide cell phone purchases
6 / 10
People addicted to their cell phones
WHY THIS MATTERS
A teen mental-health crisis episode often shows up as an ED visit and sometimes an inpatient admission -- both are claim events the insurer pays for. Youth outpatient mental-health episodes cost $2,673 per episode, and real episodes can be higher depending on follow-up care.
THE TEAM
Jeiti Trujillo
Project Manager
Maya Parra
Business Hook
Aishwarya Mundada
Lead Modeler
Mohammad Ameen
Quality Assurance
Octavio Gzain
Technical Leader
Chinmaiye Gandhi
Data Engineer
PROPOSED ANALYTICS APPROACH
BINARY CLASSIFICATION
Build a Classification Model with Decision Trees / Random Forest. High degree of explainability which is important to offer recommendations. Flag teens who cross the clinical threshold to trigger an intervention.
CLASS 1 -- HIGH RISK
Addiction_Level > 9.5
CLASS 0 -- LOW RISK
Addiction_Level <= 9.5
HYPOTHESIS A
Screen Exposure as Driver
High blue-light exposure before bed disrupts sleep cycles -- a leading physiological precursor to anxiety spikes.
VARS: Screen_Time_Before_Bed, Sleep_Hours
HYPOTHESIS B
Social Media as Driver
High-frequency checking (fragmented attention) is likely more correlated with depression than long, sustained sessions of Gaming or Education.
VARS: Phone_Checks_Per_Day, Time_on_Social_Media
Raw Data
>>>
Feature Engineering
>>>
Train/Test Split
>>>
Random Forest
>>>
Threshold @ 0.40
>>>
Risk Flag
DEFINING SUCCESS
PRIMARY GOAL
Reduce crisis-related claims -- fewer hospitalizations, emergency therapy sessions, healthier teens
PREVENTATIVE SHIFT
Increase utilization of low-cost preventative care: counseling apps, wellness coaching triggered by early alerts
PREDICTIVE PRECISION
Achieve 90%+ recall to catch high-risk cases with 14-30 day lead time before clinical crisis emerges
CRITICAL CONTEXT
In mental health prediction, false negatives carry far greater costs than false positives. Missing a high-risk teen leads to crisis; flagging a healthy teen enables preventative support -- a worthwhile investment.
ECONOMIC IMPACT METRICS
Cost Avoidance
Prevent acute interventions through early action
Lifetime Value
Improved long-term outcomes reducing chronic claims
Engagement Rate
Parents take action after receiving alerts