Cost predictions at quartile measures of quality: Summed events measure of quality.

Context

Our target was to predict gender, age and emotion from audio. We found audio labeled datasets on Mozilla and RAVDESS. So by using R programming language 20 statistical features were extracted and then after adding the labels these datasets were formed. Audio files were collected from "Mozilla Common Voice" and “Ryerson AudioVisual Database of Emotional Speech and Song (RAVDESS)”.

Content

Datasets contains 20 feature columns and 1 column for denoting the label. The 20 statistical features were extracted through the Frequency Spectrum Analysis using R programming Language. They are: 1) meanfreq - The mean frequency (in kHz) is a pitch measure, that assesses the center of the distribution of power across frequencies. 2) sd - The standard deviation of frequency is a statistical measure that describes a dataset’s dispersion relative to its mean and is calculated as the variance’s square root. 3) median - The median frequency (in kHz) is the middle number in the sorted, ascending, or descending list of numbers. 4) Q25 - The first quartile (in kHz), referred to as Q1, is the median of the lower half of the data set. This means that about 25 percent of the data set numbers are below Q1, and about 75 percent are above Q1. 5) Q75 - The third quartile (in kHz), referred to as Q3, is the central point between the median and the highest distributions. 6) IQR - The interquartile range (in kHz) is a measure of statistical dispersion, equal to the difference between 75th and 25th percentiles or between upper and lower quartiles. 7) skew - The skewness is the degree of distortion from the normal distribution. It measures the lack of symmetry in the data distribution. 8) kurt - The kurtosis is a statistical measure that determines how much the tails of distribution vary from the tails of a normal distribution. It is actually the measure of outliers present in the data distribution. 9) sp.ent - The spectral entropy is a measure of signal irregularity that sums up the normalized signal’s spectral power. 10) sfm - The spectral flatness or tonality coefficient, also known as Wiener entropy, is a measure used for digital signal processing to characterize an audio spectrum. Spectral flatness is usually measured in decibels, which, instead of being noise-like, offers a way to calculate how tone-like a sound is. 11) mode - The mode frequency is the most frequently observed value in a data set. 12) centroid - The spectral centroid is a metric used to describe a spectrum in digital signal processing. It means where the spectrum’s center of mass is centered. 13) meanfun - The meanfun is the average of the fundamental frequency measured across the acoustic signal. 14) minfun - The minfun is the minimum fundamental frequency measured across the acoustic signal 15) maxfun - The maxfun is the maximum fundamental frequency measured across the acoustic signal. 16) meandom - The meandom is the average of dominant frequency measured across the acoustic signal. 17) mindom - The mindom is the minimum of dominant frequency measured across the acoustic signal. 18) maxdom - The maxdom is the maximum of dominant frequency measured across the acoustic signal 19) dfrange - The dfrange is the range of dominant frequency measured across the acoustic signal. 20) modindx - the modindx is the modulation index, which calculates the degree of frequency modulation expressed numerically as the ratio of the frequency deviation to the frequency of the modulating signal for a pure tone modulation.

Acknowledgements

Gender and Age Audio Data Souce: Link: https://commonvoice.mozilla.org/en Emotion Audio Data Souce: Link : https://smartlaboratory.org/ravdess/

This dataset presents information from 2016 at the household level; the percentage of households within each Index of Household Advantage and Disadvantage (IHAD) quartile for Local Government Area (LGA) 2017 boundaries. The IHAD is an experimental analytical index developed by the Australian Bureau of Statistics (ABS) that provides a summary measure of relative socio-economic advantage and disadvantage for households. It utilises information from the 2016 Census of Population and Housing. IHAD quartiles: All households are ordered from lowest to highest disadvantage, the lowest 25% of households are given a quartile number of 1, the next lowest 25% of households are given a quartile number of 2 and so on, up to the highest 25% of households which are given a quartile number of 4. This means that households are divided up into four groups, depending on their score. This data is ABS data (catalogue number: 4198.0) used with permission from the Australian Bureau of Statistics. For more information please visit the Australian Bureau of Statistics. Please note: AURIN has generated this dataset through aggregating the original SA1 level data (with calculated number of households/quartile) to LGA level.

This dataset includes one dataset which was custom ordered from Statistics Canada.The table includes information on housing suitability and shelter-cost-to-income ratio by number of bedrooms, housing tenure, age of primary household maintainer, household type, and income quartile ranges for census subdivisions in British Columbia. The dataset is in Beyond 20/20 (.ivt) format. The Beyond 20/20 browser is required in order to open it. This software can be freely downloaded from the Statistics Canada website: https://www.statcan.gc.ca/eng/public/beyond20-20 (Windows only). For information on how to use Beyond 20/20, please see: http://odesi2.scholarsportal.info/documentation/Beyond2020/beyond20-quickstart.pdf https://wiki.ubc.ca/Library:Beyond_20/20_Guide Custom order from Statistics Canada includes the following dimensions and variables: Geography: Non-reserve CSDs in British Columbia - 299 geographies The global non-response rate (GNR) is an important measure of census data quality. It combines total non-response (households) and partial non-response (questions). A lower GNR indicates a lower risk of non-response bias and, as a result, a lower risk of inaccuracy. The counts and estimates for geographic areas with a GNR equal to or greater than 50% are not published in the standard products. The counts and estimates for these areas have a high risk of non-response bias, and in most cases, should not be released. Housing Tenure Including Presence of Mortgage (5) 1. Total – Private non-band non-farm off-reserve households with an income greater than zero by housing tenure 2. Households who own 3. With a mortgage1 4. Without a mortgage 5. Households who rent Notes: 1) Presence of mortgage - Refers to whether the owner households reported mortgage or loan payments for their dwelling. 2015 Before-tax Household Income Quartile Ranges (5) 1. Total – Private households by quartile ranges1, 2, 3 2. Count of households under or at quartile 1 3. Count of households between quartile 1 and quartile 2 (median) (including at quartile 2) 4. Count of households between quartile 2 (median) and quartile 3 (including at quartile 3) 5. Count of households over quartile 3 Notes: 1) A private household will be assigned to a quartile range depending on its CSD-level location and depending on its tenure (owned and rented). Quartile ranges for owned households in a specific CSD are delimited by the 2015 before-tax income quartiles of owned households with an income greater than zero and residing in non-farm off-reserve dwellings in that CSD. Quartile ranges for rented households in a specific CSD are delimited by the 2015 before-tax income quartiles of rented households with an income greater than zero and residing in non-farm off-reserve dwellings in that CSD. 2) For the income quartiles dollar values (the delimiters) please refer to Table 1. 3) Quartiles 1 to 3 are suppressed if the number of actual records used in the calculation (not rounded or weighted) is less than 16. For cases in which the renters’ quartiles or the owners’ quartiles (figures from Table 1) of a CSD are suppressed the CSD is assigned to a quartile range depending on the provincial renters’ or owners’ quartile figures. Number of Bedrooms (Unit Size) (6) 1. Total – Private households by number of bedrooms1 2. 0 bedrooms (Bachelor/Studio) 3. 1 bedroom 4. 2 bedrooms 5. 3 bedrooms 6. 4 bedrooms Note: 1) Dwellings with 5 bedrooms or more included in the total count only. Housing Suitability (6) 1. Total - Housing suitability 2. Suitable 3. Not suitable 4. One bedroom shortfall 5. Two bedroom shortfall 6. Three or more bedroom shortfall Note: 1) 'Housing suitability' refers to whether a private household is living in suitable accommodations according to the National Occupancy Standard (NOS); that is, whether the dwelling has enough bedrooms for the size and composition of the household. A household is deemed to be living in suitable accommodations if its dwelling has enough bedrooms, as calculated using the NOS. 'Housing suitability' assesses the required number of bedrooms for a household based on the age, sex, and relationships among household members. An alternative variable, 'persons per room,' considers all rooms in a private dwelling and the number of household members. Housing suitability and the National Occupancy Standard (NOS) on which it is based were developed by Canada Mortgage and Housing Corporation (CMHC) through consultations with provincial housing agencies. Shelter-cost-to-income-ratio (4) 1. Total – Private non-band non-farm off-reserve households with an income greater than zero 2. Spending less than 30% of households total income on shelter costs 3. Spending 30% or more of households total income on shelter costs 4. Spending 50% or more of households total income on shelter costs Note: 'Shelter-cost-to-income ratio' refers to the proportion of average total income of household which is spent on shelter costs. Household Statistics (8) 1....

This dataset presents information from 2016 at the household level; the percentage of households within each Index of Household Advantage and Disadvantage (IHAD) quartile for Statistical Area Level 2 (SA2) 2016 boundaries. The IHAD is an experimental analytical index developed by the Australian Bureau of Statistics (ABS) that provides a summary measure of relative socio-economic advantage and disadvantage for households. It utilises information from the 2016 Census of Population and Housing. IHAD quartiles: All households are ordered from lowest to highest disadvantage, the lowest 25% of households are given a quartile number of 1, the next lowest 25% of households are given a quartile number of 2 and so on, up to the highest 25% of households which are given a quartile number of 4. This means that households are divided up into four groups, depending on their score. This data is ABS data (catalogue number: 4198.0) used with permission from the Australian Bureau of Statistics. For more information please visit the Australian Bureau of Statistics. Please note: AURIN has generated this dataset through aggregating the original SA1 level data (with calculated number of households/quartile) to SA2 level.

Feature preparation Preprocessing was applied to the data, such as creating dummy variables and performing transformations (centering, scaling, YeoJohnson) using the preProcess() function from the “caret” package in R. The correlation among the variables was examined and no serious multicollinearity problems were found. A stepwise variable selection was performed using a logistic regression model. The final set of variables included: Demographic: age, body mass index, sex, ethnicity, smoking History of disease: heart disease, migraine, insomnia, gastrointestinal disease, COVID-19 history: covid vaccination, rashes, conjunctivitis, shortness of breath, chest pain, cough, runny nose, dysgeusia, muscle and joint pain, fatigue, fever ,COVID-19 reinfection, and ICU admission. These variables were used to train and test various machine-learning models Model selection and training The data was randomly split into 80% training and 20% testing subsets. The “h2o” package in R version 4.3.1 was employed to implement different algorithms. AutoML was first used, which automatically explored a range of models with different configurations. Gradient Boosting Machines (GBM), Random Forest (RF), and Regularized Generalized Linear Model (GLM) were identified as the best-performing models on our data and their parameters were fine-tuned. An ensemble method that stacked different models together was also used, as it could sometimes improve the accuracy. The models were evaluated using the area under the curve (AUC) and C-statistics as diagnostic measures. The model with the highest AUC was selected for further analysis using the confusion matrix, accuracy, sensitivity, specificity, and F1 and F2 scores. The optimal prediction threshold was determined by plotting the sensitivity, specificity, and accuracy and choosing the point of intersection as it balanced the trade-off between the three metrics. The model’s predictions were also plotted, and the quantile ranges were used to classify the model’s prediction as follows: > 1st quantile, > 2nd quantile, > 3rd quartile and < 3rd quartile (very low, low, moderate, high) respectively. Metric Formula C-statistics (TPR + TNR - 1) / 2 Sensitivity/Recall TP / (TP + FN) Specificity TN / (TN + FP) Accuracy (TP + TN) / (TP + TN + FP + FN) F1 score 2 * (precision * recall) / (precision + recall) Model interpretation We used the variable importance plot, which is a measure of how much each variable contributes to the prediction power of a machine learning model. In H2O package, variable importance for GBM and RF is calculated by measuring the decrease in the model's error when a variable is split on. The more a variable's split decreases the error, the more important that variable is considered to be. The error is calculated using the following formula: 𝑆𝐸=𝑀𝑆𝐸∗𝑁=𝑉𝐴𝑅∗𝑁 and then it is scaled between 0 and 1 and plotted. Also, we used The SHAP summary plot which is a graphical tool to visualize the impact of input features on the prediction of a machine learning model. SHAP stands for SHapley Additive exPlanations, a method to calculate the contribution of each feature to the prediction by averaging over all possible subsets of features [28]. SHAP summary plot shows the distribution of the SHAP values for each feature across the data instances. We use the h2o.shap_summary_plot() function in R to generate the SHAP summary plot for our GBM model. We pass the model object and the test data as arguments, and optionally specify the columns (features) we want to include in the plot. The plot shows the SHAP values for each feature on the x-axis, and the features on the y-axis. The color indicates whether the feature value is low (blue) or high (red). The plot also shows the distribution of the feature values as a density plot on the right.

Affordability ratios calculated by dividing house prices by gross annual workplace-based earnings. Based on the median and lower quartiles of both house prices and earnings in England and Wales.

AL refers to the axial length, CCT to the central corneal thickness, ACD to the external phakic anterior chamber depth measured from the corneal front apex to the front apex of the crystalline lens, LT to the central thickness of the crystalline lens, R1 and R2 to the corneal radii of curvature for the flat and steep meridians, Rmean to the average of R1 and R2, PIOL to the refractive power of the intraocular lens implant, and SEQ to the spherical equivalent power achieved 5 to 12 weeks after cataract surgery.

Tracks the daily sea ice extent for the Arctic Circle and Antarctica using the NSIDC's Sea Ice Index dataset, as well as pre-calculating several useful measures: historical inter-quartile range across the year, the previous lowest year and the previous year.

Model 1: adjusted for age at follow-up, gender, intervention group.Model 2: as model 1 plus adjustment for z-score of birth weight, father's social class, lifetime smoking, alcohol intake and exercise.1Insulin Sensitivity Index whilst fasting = 104/(I0×G0).2Corrected Insulin Response at 30 minutes = 100×I30/(G30×(G30−70).†Outcomes were natural-log transformed, and coefficients and confidence intervals represent a change in ratio of geometric means per quartile of formula/cows' milk intake.*Reference category is those in the lowest quartile of infant formula/cow's milk intake, amongst those who received infant formula/cow's milk.

Data were presented as means with SDs or number with percentage. MAP, mean arterial pressure; PaCO2, partial pressure of carbon dioxide; PaO2, partial pressure of oxygen; BUN, blood urea nitrogen; AST, aspartate transaminase; ALT, alanine transaminase.*Plasma PQ concentration performed in 79 cases out of a total of 136 patients.

Differences between lower and upper quartiles of scales.

Because we didn't have data in the nominal groups that Lesnick et al. defined (our predicted probabilities ranged from 0.328 to 0.634), we did not use the Lesnick et al. groupings. To be able to calculate odds ratios we broke the predicted probabilities into quartiles and calculated the odds ratios for each of the groups.*Wald test with 3 degrees of freedom.

Baseline characteristics of the study population according to CMI quartiles.

WHO age-standardized prevalence estimates for single long-term conditions: Malawi, The Gambia and Uganda.

Baseline lifestyle factor prevalence estimates: Malawi and Uganda.

Adjusted associations between average levels of inflammatory biomarkers by quartiles of distribution and odds of LEAF.

Table showing the average age, compressed breast thickness and volumetric breast density of the patients. The volumetric breast density has been assessed by Volpara Solutions. The value provided is the mean with the standard deviation in brackets. The overall average breast density top quartile value was used across the study.