| Popis: |
This paper analyzes a data set obtained from a recent study per ormed at The Karolinska Institute. The data set is comprised of 131 children with anxiety disorders, aged 8 - 12, who all underwent a novel treatment against their disorder called internet-delivered cognitive behavior therapy (ICBT). The data set contains standardized clinical severity ratings (CSR) of the patients before and and after the treatment, as well as 233 features for each patient (demographicinformation, symptom reports, information on other diagnoses etc). Before thetreatment, the clinicians also made a "guess", scored on a scale of 1 - 10, answering the question"How successful will ICBT treatment be for this particular patient?". Firstly, this studyfound that the clinicians predicted remission with an accuracy of approximately 50%. Secondly,this study employed machine learning algorithms designed to learn from the data setand make predictions based on the feature information of each particular patient. The topperforming algorithm predicted with an accuracy of 70%. This study therefore suggests thatmachine learning algorithms can predict outcome of ICBT treatmentwith a higher level of accuracythan clinicians. This study then addresses its weaknesses and limitations to this conclusion,most importantly the vagueness of the question and scale that the clinicians basedtheir guesses on. |
