False Flags No More: Improving Accuracy for Alatrac’s Anomaly Detection Model

Industry

Technology

Challenge

Alamon, Inc. faced challenges with their Alatrac platform’s AI anomaly detection, which produced excessive false positives and missed critical issues. Inspectors often had to revisit sites, causing costly delays. With 180M+ data points and human variability, Alamon needed a refined AI model to enhance accuracy and efficiency.

Results

The enhanced AI model delivered significant improvements: False Positives Reduction – Alatrac's anomaly detection rate plummeted from 50% to 0.73%, representing a 98.54% improvement. Real-Time Detection – Issues are now flagged immediately during inspections, preventing costly rework and operational delays. Efficiency Boost – Less rework & oversight saved time/resources. Business Impact – Alamon leads with high data accuracy for compliance-driven clients.

Solution

Data Pre-processing, Feature Engineering, Model Optimization, Weekly Retraining

alamon.com

Milvian Group transformed our approach to anomaly detection. Their expertise turned a flawed model into a reliable system, vastly improving our data accuracy and operational efficiency.

Travis Hansen

Product Manager @ Alamon, Inc.

Alamon, Inc

Alamon was established in 1975 by Peg and Frank Gebhardt as a telecommunications installation company. Today, it is a thriving employee-owned company with a broader mission. Alamon prioritizes delivering the highest quality service while ensuring fair wages and benefits for its talented workforce.

The Challenge

Alamon, Inc. faced significant challenges with their proprietary Alatrac platform, which is used throughout the Utility Industry for the inspections of over 500,000 utility pole inspections annually. While the platform was designed for efficient data collection and compliance, it was difficult to assign a confidence rating to individual inspections and track inspectors that weren't completing tasks effectively.

Based on these challenges, it is critical for Alatrac to be able to isolate anomalies and address issues and ensure inspections are done correctly. With over 180 million data points and diverse human behavior influencing the data, Alamon struggled with costly errors, inefficiencies, and most importantly raising the confidence of inspections to ensure they were completed correctly. While Alamon had an existing baseline model to isolate anomalies, they were getting inaccurate information and excessive false positives which could have proved costly from a re-inspection standpoint. A solution was needed to improve accuracy and prevent costly errors and potential liability for Alatrac and their customers.

The Solution

To address sthese challenges, Alamon chose to partner with Milvian Group due to their proven expertise in data preparation and AI model training and optimization. Together, they were able to identify key data points and then retain the model, enhancing the AI capabilities for accuracy. Leveraging their expertise in data science and AI, Milvian Group undertook:

Data Pre-processing: Eliminating false flags by refining the dataset, focusing on valid inspection events, and cleaning test/demo data.

Feature Engineering: Creating critical metrics, such as inverse inspection time and distance correlations, for a more comprehensive analysis.

Model Optimization: Implementing the Isolation Forest algorithm within AWS SageMaker to improve detection precision.

Automated Weekly Model Retraining: Automating data extraction and model retraining to ensure continuous improvement.

This collaborative process included iterative testing and validation with Alamon’s team, ensuring alignment between technical outputs and operational needs.

The Results

The enhanced AI model delivered significant benefits:

Reduction in False Positives: Alatrac’s overall anomaly detection rate reduced from 50% to 0.73% (a 98.54% improvement)

Real-Time Detection: The system now flags issues during inspections, preventing costly rework and delays.

Increased Efficiency: The model significantly reduced the need for both re-work, and manual oversight, saving time and resources.

Business Impact: Alamon retains its competitive advantage in the AI-driven GIS market by providing high data accuracy, a critical factor for Almon’s compliance-driven customers.