1 An Explainable Machine Learning Model for the Prediction of Parkinson’s Disease using LIME on Speech Signals (Draft Version) Richard Delwin Myloth Pavan Rajkumar Magesh Rijo Jackson Tom Dept. of Computer Science and Engineering CMR Institute of Technology {rich17cs, pava17is, rijo.j}@cmrit.ac.in Abstract Speech impairments analysis has been used as an efficient tool for early detection of Parkinson’s disease (PD). In this paper we explore explainable models for the proposed Neural Network, SVM and Random Forest model developed to classify PD patients from the non-Pd patients utilising the speech samples. The performance of the method has been assessed with a reliable dataset from UCI repository. The proposed achieves an accuracy of 71.6%, 74.5% and 68.75% for the neural network, random forest and the SVM model respectively. Explainable models for the neural network and SVM have been generated with the help of LIME. The obtained results are compared with the results of the random forest model obtained by backtracking its decision path to obtain the most contributing features. The results are compared and its necessity are studied for explainable decision support in medicine. Introduction The World Health Organization (WHO) depicted neurological disorders as one of the major threats to public health. Parkinson’s disease (PD), stroke, multiple sclerosis, headache disorders, dementia, epilepsy, etc. are amongst the most common disorders. Parkinson’s disease (PD), initially called shaking palsy, first was described by James Parkinson in 1817[1]. Parkinson's disease is a progressive degenerative disorder of the basal ganglia that affects the initiation and execution of voluntary movements. It is the second most common neurodegenerative disorder after Alzheimer's disease.[2] The symptoms of Parkinson's usually begin gradually and get worse over time. Furthermore, the prevalence of PD is going to increase due to the aging population. There is no definitive test for the diagnosis of PD, the disease must be diagnosed based on clinical criteria. Rest tremor, bradykinesia, rigidity, and loss of postural reflexes are generally considered the cardinal signs of PD[2][3]. Apart from these symptoms of the patients, currently, numerous physical tests such as (MRI, PET), etc are used to determine the plausibility of Parkinson's which are centred around assessing the dopamine levels in the brain. Although progressive parkinsonism which is referred to as Parkinson’s Disease, can be diagnosed upfront in patients with typical presentations of the above-mentioned cardinal signs, the differential diagnosis versus other forms of parkinsonism can be challenging, especially early in the disease when signs and symptoms of different forms of parkinsonism have greater overlap.[2] Based on recent research findings, PD is much more than degeneration of the dopaminergic nigrostriatal system; the first neurons affected in PD are nondopaminergic. In addition to this population-based investigation have shown that no less than 15% of patients diagnosed with PD in the population do not satisfy strict clinical criteria for the disease, and roughly 20% of patients with Parkinson’s disease who have already received medical attention have not been diagnosed with the disease. Thus, by the time the disease has been diagnosed, around 60% of the nigrostriatal neurons degenerated, and 80% of striatal dopamine is depleted.[4] Apart from the four cardinal symptoms, speech disorders, visual hallucinations, etc. are few of the abnormalities reported according to [5]. Speech disorders in patients with PD are characterized by monotonous, soft, and breathy speech with variable rate and frequent word-finding difficulties. Collectively, these speech symptoms are called hypokinetic dysarthria and it affects 90% of its patients (of over 50 years of age).