Presently a majority of the efforts to respond to risks from lead paint in housing have concentrated on retrospective-based actions relying on the detection of elevated blood lead levels in children residing in target housing. The failure of this approach is evidenced by the over 1.2 million children suffering from lead poisoning- poisoning that results in brain damage, slowed development, hearing, speech and learning problems. Furthermore, such retroactive-oriented programs fail to assess the true dimensions of the potential risk. Over 80% of housing built before 1978 is estimated to contain lead paint. An effective proactive prevention program is needed to meet the potential risk from this widespread use of lead paint in housing.
The basic premise of the proposed project is that advances in mobile communication applications, image recognition and online training now make it possible to create much more effective inspection programs at a greatly reduced cost. The expertise of experts can be multiplied and the guidance to field inspectors can be greatly strengthened using existing technology, without increasing inspection costs..
To develop a proactive program to characterize residential lead risk, models and tools need to be developed to raise the ability of general or code building inspectors to make predictive residential lead risk evaluations. At present general inspectors have been given little guidance and few tools to make reliable risk judgments. EPA-accredited Risk experts known as lead risk assessors are capable of making such judgements, but their widespread use as inspectors has proven too expensive an investment. The objective of this project is to develop predictive analytic approaches using mobile applications and artificial intelligence to enhance the capacity of building inspectors to make reliable risk judgments. It is no longer necessary for an accredited risk expert to physically visit a residential site to make an assessment. Communication technology and artificial intelligence allows general inspectors or even informed residents, using feedback and online communication with risk assessor experts, to access the guidance needed to remotely assess lead risks.
The project entails four phases:
- Data collection
- Model development
- General Inspector training
- Post training support
The first phase of the project involves collection of data on a number of residential structures by accredited risk assessment experts. The initial phase would involve 100 inspections. These inspections would be carried out by experienced assessors and collect such data as the age of the building, the existence of bare soil surrounding the building, the room types within the building and their uses in accordance with existing protocols, as well as photos taken in a variety of settings to give evidence of the condition of the paint in the residence supported by XRF sampling. The data would be collected using software on a mobile device and uploaded into a central database.
The second phase of the data collection effort involves the examination of the inspection data by 4 to 6 risk experts. For each residence inspected, the expert risk assessor would make a judgment on the condition of the paint shown in the inspection photographs and assign each residence an overall lead risk number from 1 to 5.
A third phase of data collection will take place after general inexperienced inspectors are trained online. The results of the collection of inspection data taken by non-expert inspectors will be recorded and stored in a central database to compare to expert judgements in subsequent phases of the project.
Using the expert risk assessments as dependent variables and the residential condition inspection data as independent variables, a statistical model will be developed using logit analysis. This analysis should reveal the statistical significance of different residential condition variables on the expert opinion of the level of lead risk.
A second model will be explored to categorize the condition of paint based on photographic analysis. The expert judgement data of the condition of the residential paint based on inspection photographs will be employed in image recognition software to assess whether paint conditions can be reliably assessed by image recognition software using inspection data.
An online training course will be developed based on the findings of the preliminary data collection and modeling efforts. The first objective of the course is to train the inspectors in the use of the data collection software. Such software must have the capacity to record numerical, text and sample data, as well as photographic images. Simulated inspections and software guidance will be employed as well as field experience to ensure the general inspectors or other home visitors can both use the software and upload data to the central database.
The second phase of the course will discuss how to collect the residence building variables found to be significant in the statistical model. This phase involves more than simply listing the data elements, but includes the challenges and obstacles facing an inspector during an inspection. Part of this phase will also involve providing training on how to make judgements regarding paint conditions using existing online approaches used by HUD.
Post training support and validation
The initial training will be reinforced by feedback between the risk assessment expert trainers and the inspectors subsequent to the course. The experts will be able to view the inspection data and photos of subsequent inspections by the trainees and the trainees will be able to ask questions of the experts via the mobile app software. This will provide economical expert calibration and allow for immediate comment on trainee judgements. In addition, the database will expand based on the inspectors’ activities. The models can be recalibrated at very little cost and new sources of risk identified.
A second validation of the approach might be made by augmenting the database with the housing characteristics of housing in which elevated blood levels were found. Analysis of a sample of such housing would provide an assessment of the validity of the expert judgement assessments.
The approach of using mobile application software, online training and artificial intelligence software has the potential to create enhanced housing risk assessments beyond the lead paint issue. The development of the proposed approach could be replicated for other housing risks eg mold, asbestos, safety hazards. The basic premise is that risk expert resources can be multiplied by enhance training and communication with inspectors and residents. By the use of existing technologies , a program of an integrated risk “continuity of care” for building stock could replace the present fragmented and periodic approaches.