Evaluating the Accuracy of Supervised Machine Learning in Predicting Dental Treatment Durations: An Experimental Study

In recent years, the integration of technology in healthcare has revolutionized the way treatments are administered and managed. A pivotal aspect of this transformation lies in the use of machine learning (ML) to enhance clinical decision-making. This experimental study aims to assess the accuracy of supervised ML in predicting treatment durations at multiple dental hospitals and clinics, while also evaluating how these predictions impact clinical workflow efficiency.

Ethical Considerations

This research has been meticulously designed to adhere to ethical standards governing clinical studies. The Ethics Committee at the College of Dentistry, University of Sulaimani approved the research project under Code No. (COD-EC-25-0077) on March 24, 2025. Ensuring participant safety and compliance with relevant guidelines is paramount in this study.

Inclusion Criteria

To facilitate comprehensive results, the study encompasses a diverse group of participants. It includes patients over the age of 18 undergoing common dental treatments such as:

• Fillings
• Root canals
• Periodontal treatment
• Tooth preparation
• Implant visits
• Orthodontic treatment
• Extractions

Younger patients under 18 were also considered, with informed consent obtained from their parents. This broad inclusion allows for a varied data pool which is vital for the training of the machine learning models.

Exclusion Criteria

Conversely, several parameters were established to maintain the integrity of the study. Patients with incomplete records, emergency cases requiring immediate intervention, or those undergoing complex or rare procedures were excluded from participation. This decision was necessary to ensure that the data used for analysis was both relevant and reliable.

Methodology

Patient Population and Data Collection

The study involved the collection of data from a total of 2,500 patients, which served as a training set for the machine learning model. An additional 250 cases were utilized for validation, comparing the model’s predictions against actual treatment durations. The patient selection process is elegantly illustrated in a flow diagram (Fig. 1), detailing the methodology from recruitment to data analysis.

Machine Learning Model Description

The core of this study is a hybrid machine learning system designed to predict dental procedure durations. This system fuses real-time online data retrieval with clinical expertise, ensuring enhanced accuracy and efficiency.

Machine Learning Core

The prediction engine employs a two-tiered modeling strategy using the sklearn library’s Linear Regression function. Key features of the model include:

• Numerical Features: Dentist experience (in years), patient age.
• Categorical Features: Patient sex and dentist specialty (one-hot encoded).

To manage sparse data effectively, a lookup table fallback mechanism is integrated. This ensures that procedures with inadequate records utilize precomputed averages, minimizing the risk of unreliable model outputs.

Efficiency Optimizations

In improving the model’s operational efficiency, several strategies were employed:

1. Column Alignment: Dynamically reindexing prediction inputs ensures readiness for training without the overhead of replaying the full dataset.

2. Memory Management: By limiting DataFrame construction during real-time prediction, resource allocation is optimized.

3. Parallel Training: Independent model construction for each procedure enables scalable adaptations within the healthcare setting.

Online Data Retrieval Module

To augment predictions with real-time data, the system integrates live dental guideline searches. Utilizing an API to retrieve relevant information allows for precise estimates of treatment durations, enhancing the model’s predictive capability.

Clinical Safety Mechanisms

To maintain safety and reliability, the model incorporates various safety checks:

• Procedure Minimums Dictionary: Establishes biologically plausible timeframes.
• Specialty-Adjusted Predictions: Factors in specialists’ efficiencies, providing context-sensitive predictions.

User Interface and Workflow Integration

The study’s prediction software features a user-friendly graphical interface developed using Tkinter (Fig. 2). This interface involves:

• Dynamic Form Validation: Ensuring inputs are accurately completed in real time.
• Progressive Disclosure: Tailoring elements based on system readiness to streamline user interaction.
• Comparative Insights: Delivering simultaneous views of model-generated and online estimates.

Computational Performance

The system’s performance is noteworthy. Benchmarked against an Intel i7-1185G7 processor, model training takes between 120 and 450 milliseconds per procedure, depending on sample size. Prediction latency averages less than 15 milliseconds for dedicated models, and under 2 milliseconds for fallback queries. Remarkably, the total memory footprint remains below 45 MB even with a substantial dataset of 10,000 procedure records.

Validation Framework

Three-tier validation ensures the model’s accuracy and reliability:

1. Input Sanitization: Validating numerical input types and ranges prevents data integrity issues.
2. Model Confidence Checking: Mechanisms that trigger fallback predictions when model uncertainty surpasses predefined thresholds.
3. Clinical Plausibility Gates: Constraining predictions to biologically feasible timeframes, reinforcing the model’s credibility.

Data Collection and Analysis

Data acquisition will primarily involve manual entries recorded by clinicians across various dental clinics. The information gathered will include:

• Actual treatment duration as documented by dental professionals.
• Types of dental procedures performed.
• Specialists’ expertise, categorized by experience levels.
• Demographics such as patient age and gender.

To prepare the dataset, strategies for handling missing data include rigorous manual preprocessing to standardize treatment classifications and correct any typographical errors.

Statistical Analysis

The gathered data will undergo thorough statistical analysis using IBM SPSS and other analytical tools. Key metrics employed will include:

• Descriptive Statistics: Summarizing both predicted and actual treatment durations.
• Paired t-tests: Assessing the significance of differences based on gender, age, dentist experience, and specialty.
• R² Score and Mean Absolute Error (MAE): Evaluating model performance in predicting treatment durations.

A significance threshold of a p-value less than 0.05 underscores the robustness of the findings, facilitating reliable clinical implications from the study.

This meticulously structured research endeavors to demonstrate not only the feasibility of utilizing ML for enhancing healthcare efficiency but also the necessity for ethical and practical considerations in employing technology within clinical settings. As dentistry continues to evolve, studies like this will pave the way for smarter, more effective healthcare services.